Site Project Catalog — Tilden Mine¶
Purpose: Groups Tilden's 54 initiatives (TLD-01..TLD-54) into actionable projects for site leadership. Each site project bundles related initiatives, sizes the local opportunity, and references the corporate project it feeds into.
Audience: Site leadership, Keith Holmgren (Sr Director Mining Technology), internal team
Important: Tilden is the only mine in the engagement. Many projects are mining-specific with no steel-site parallel. Where corporate projects (PRJ-01..08) apply, they are reframed for mining context. Mining-specific projects are new additions to the portfolio.
Last updated: 2026-04-16 (consistency pass: corporate project mappings verified against ch5/ch8. TLD-P12 → PRJ-09)
Project Summary¶
| ID | Project | Horizon | Corporate | Bundled Initiatives | Value ($/yr) | Champion | Status |
|---|---|---|---|---|---|---|---|
| TLD-P01 | Concentrator Desliming & Recovery Optimization ★★★ (POV) | H1 | PRJ-10 | TLD-23, TLD-08, TLD-32, TLD-54 | $8-18M | Keith Holmgren, Dan McGrath, Sean Halston | POV confirmed ★★★ Apr 7 |
| TLD-P02 | Concentrator Grinding & Instrumentation | H2 | PRJ-04 (reframed) | TLD-07, TLD-22, TLD-11 | $4-9M | Dan McGrath, Sean Halston, Todd Davis | identified (grinding deferred Apr 7) |
| TLD-P03 | Pellet Plant Quality & Control | H1→H2 | PRJ-04 (reframed) | TLD-09, TLD-34, TLD-10 | $3-9M | Process engineering team | identified |
| TLD-P04 | Mining Fleet PdM & Lifecycle Intelligence ★★★ | H1→H3 | PRJ-03 | TLD-02, TLD-19, TLD-46, TLD-47 | $9-24M | Pete Austin, Chase Lincoln | validated ★ Stepping stone |
| TLD-P05 | Fixed Plant PdM & Failure Analytics | H1→H2 | PRJ-03 + PRJ-01 | TLD-03, TLD-42, TLD-41 | $4-12M | George Harmon, Gary | identified |
| TLD-P06 | Drill & Blast Intelligence ★★★ | H2 | new (mining) | TLD-05, TLD-53 | $1-4M | Jeff Domann | identified |
| TLD-P07 | Mine Operations & Dispatch Intelligence ★★★ | H2→H3 | PRJ-07 + PRJ-02 (reframed) | TLD-04, TLD-15, TLD-50, TLD-26 | $6-20M | Brad Koski, Tyler Craig, Molly | identified ★ Ryan #2 |
| TLD-P08 | Mine-to-Dock Logistics Optimization ★★★ | H1→H2 | PRJ-07 (reframed) | TLD-16, TLD-37 | $2-6M | Kevin (train scheduling) | validated ★ Ryan #3 |
| TLD-P09 | Ops-Maintenance Data Integration | H1 | PRJ-01 | TLD-01, TLD-45, TLD-49 | $4-9M | Pete Austin, George Harmon | identified |
| TLD-P10 | HPGR Knowledge Base + PdM Pilot ★★★ | H1 | PRJ-06 + PRJ-03 | TLD-38, TLD-33 | $2-5M | Adam Bingham | validated ★ Stepping stone |
| TLD-P11 | Maintenance Workflow & Inventory Intelligence | H1 | PRJ-06 | TLD-12, TLD-35, TLD-48, TLD-30, TLD-13 | $3-10M | Adam Bingham, warehouse team | identified |
| TLD-P12 | Mining Knowledge Capture & Virtual SME | H1→H2 | PRJ-09 | TLD-14, TLD-43, TLD-51, TLD-52, TLD-27 | $2-7M | Adam Bingham, Lynn Casco, Brad Koski | identified |
| TLD-P13 | Maintenance Planning & Scheduling | H1→H2 | new | TLD-39, TLD-40, TLD-36 | $2-7M | Gary, Pete Austin, JR | identified |
| TLD-P14 | Workplace Safety & Inspection Digitization | H1 | new (MDT-P07 parallel) | TLD-24, TLD-20, TLD-25 | $1-4M | Dan Clarendon | identified |
| TLD-P15 | Environmental, Utilities & Geotechnical | H2 | new | TLD-18, TLD-28, TLD-17 | $1-4M | Brent (environmental) | seed |
| TLD-P16 | HR & Administrative Operations | H1 | new | TLD-29, TLD-44 | $0.3-1M | Lynn Casco | seed |
| TLD-P17 | Mine-to-Concentrator Ore Intelligence | H2 | new (mining) | TLD-21, TLD-06, TLD-31 | $8-18M | Todd Davis, Dan Collins, + mine eng TBD | validated ★ (deferred — "two or three projects down the road") |
Total addressable value: $50-160M/yr
Status key: seed = grouped from initiatives, not yet validated | identified = evidence from workshops, not yet validated with leadership | validated = leadership agrees this is a real project | sized = value estimate refined | approved = selected for roadmap
Project Cards¶
TLD-P01: Concentrator Desliming & Recovery Optimization ★★★ POV¶
| Field | Value |
|---|---|
| Horizon | H1: Bridge the Gap |
| Corporate project | PRJ-10 — Process Chemistry Optimization |
| Status | POV confirmed ★★★ — Day 5 Readout (Ryan's #1) → Mar 28 working session (scope refined) → Apr 7 IE×Tilden (POV scope locked: desliming + flotation recovery). IE writing project charter. Follow-up Thu/Fri Apr 10-11. |
| Champion(s) | Keith Holmgren (Sr Director Mining Technology, 32 yrs at Cliffs — THE concentrator SME, bus factor risk), Dan McGrath (process engineer, metallurgical), Sean Halston (ECSM concentrator ops), met tech team (TBD — Ryan has 90% of project team identified) |
| IE lead | Erico Lemos (project manager, writing charter), Bob Zadel (IE president, senior technical advisor — worked at Tilden in 1990s on flotation optimization) |
Local problem statement:
The concentrator desliming circuit is the most operator-dependent variable in Tilden's entire value chain. Metallurgical technicians (met techs) manually key in 4 dispersant (polyacrylic acid / PAA) numbers per section, controlling how much tailings are rejected in the D-slime thickeners. PAA chelates hardness ions (calcium/magnesium) that would otherwise non-selectively flocculate particles, acting as a "throttle" on weight rejection — from ~10% (zero PAA) to ~50% (max PAA), with the optimum around 30-35%. The relationship between PAA dose and weight rejection varies with ore type, pH, water temperature, and mineralogy.
Between 6-hourly metallurgical balances, met techs take 1-liter beaker settling tests — a purely visual, subjective measurement where the same 30% weight rejection can look like 110 mils or 90 mils depending on ore characteristics. Experienced operators (28+ years) calibrated this judgment over decades. They are being replaced by staff with 2 years of experience and half a week of training who "don't even know what questions to ask."
If desliming is under-dispersed, the flotation circuit gets overwhelmed and can literally flood the plant. The operator bias is conservative: "don't flood the plant" takes priority over optimizing recovery. This leaves value on the table.
The concentrator goes the way the desliming goes. Tilden goes the way the concentrator goes. (Ryan, Apr 7)
Scope decisions (Apr 7): Grinding explicitly OUT of scope per Keith — "the grinding circuit is outside of scope and our D-slime and flotation section is the customer of the grinding circuit." Water chemistry OUT of scope per Keith — "more of a course longer-term control," but referenced as data input. Mine-to-concentrator ore tracking DEFERRED per Ryan — "two or three projects down the road" (moved to TLD-P17).
Bundled initiatives: - TLD-23: Reagent Suite Optimization / Dispersant Standardization ★★★ — Track 1. Data-driven dispersant dosing per section. Correlate met tech measurements + process variables (tailing sump levels, DTU pump speeds, thickener profiles) + metallurgical balance outcomes to recommend optimal PAA adjustments. Goal: every met tech response as consistent as the best 35-year veteran's. - TLD-08: Flotation Recovery Optimization ★★★ — Downstream benefit. Better desliming → more stable flotation feed → reduced amine consumption → improved iron recovery. The primary Y metric for the POV. Ryan: 0.5% weight recovery = 100K tons = tens of millions of dollars. - TLD-32: Concentrator Operator Decision Support / Live CRP ★★★ — Keith's vision: a "live control and response plan" that evaluates current process state, identifies bottleneck, and provides actionable recommendations. Replaces static Word CRPs that nobody reads. Daily management board for process engineers. - TLD-54: Beaker Test Vision & Standardization ★★ — NEW. Camera-based standardization of beaker settling tests at autocompo PSI stations. Removes operator-to-operator measurement variability. Feeds objective data into dispersant model. Keith conceived, Bob Zadel validated feasibility.
Systems involved: DCS (Foxboro IA), Pi historian (1.3B entries), D-slime grab sample records (Kevin's spreadsheet — 4 samples/day), Courier silica analyzer (7-min cycle), autocompo PSI stations (one per section), reagent dosing logs, flotation cell instrumentation, tailing sump level sensors, DTU pump speed data
Value estimate: $8-18M/yr - Reagent savings (5-10% of $50M) = $2.5-5M - Recovery improvement (each 1% weight recovery = ~100K tons at $100+/ton) = $4-10M - Throughput stability from reduced reactive adjustments + reduced flotation flooding events = $1.5-3M
Confidence: Very High — POV-level commitment from site leadership (Ryan, Keith), IE leadership (Bob Zadel, Erico), and corporate (Chad directed concentrator focus). Keith provided deepest technical walkthrough to date. Data confirmed to exist (Kevin's spreadsheet, Pi, DCS).
Implementation approach (8-week POV): 1. Weeks 1-2: Data assessment + workflow mapping — Validate dispersant data availability (DCS/Pi vs. paper). Map met tech workflows with 2-3 experienced met techs. Capture Keith's bottleneck classification framework. Begin D-slime grab sample data analysis. IE operational integration: ride-alongs, shift handoff practices, current CRP review. 2. Weeks 3-5: Model development — Build desliming optimization model correlating process variables to weight rejection outcomes. Historical backtest on known events. Prototype beaker vision if camera setup feasible. 3. Weeks 6-7: Validation + operator acceptance — Met tech validation of model recommendations. Live advisory testing on 1-2 sections. Operator feedback loop. Daily management board prototype. 4. Week 8: Deliverables — Desliming optimization model, data quality scorecard, live CRP prototype, operational integration design (IE), process optimization roadmap with measured value.
Dependencies: Keith Holmgren availability (bus factor — key SME), met tech cooperation for workflow mapping, DCS/Pi data access (IT/OT security process), Kevin's D-slime grab sample data
Palmer readout alignment: - Scalability: methodology (data-driven process control standardization) applicable to any CLF concentrator/process - Quick-ROI: desliming optimization delivers measurable value within POV (weight recovery baseline → improvement) - Chad directed: concentrator was Chad's explicit directive at Mar 24 corporate readout
Apr 7 action items: - [ ] Erico writing project charter (Y = weight recovery, X's = process variables) — follow-up Thu/Fri Apr 10-11 - [ ] Keith to provide historical D-slime grab sample data (Kevin's spreadsheet) - [ ] Bob wants weight recovery baseline (mean + variability) for charter - [ ] Ryan & Keith to finalize project team roster (90% identified, not yet notified) - [ ] Confirm whether dispersant numbers are logged in DCS/Pi or paper only - Palmer named: no specifically — mining-specific. But $50M reagent anchor and process optimization align with his criteria.
★ This is Tilden's version of the information flow problem. At Cleveland, the gap was between operations and maintenance. At Middletown, between finishing lines and quality data. At Tilden, it's between the pit (ore quality data) and the concentrator (process response). Same thesis, different manifestation.
TLD-P02: Concentrator Grinding & Instrumentation¶
| Field | Value |
|---|---|
| Horizon | H2: Build the Foundation |
| Corporate project | PRJ-04 — Through-Process Quality & Yield (reframed for mining) |
| Status | identified — Grinding explicitly deferred by Keith (Apr 7): "the grinding circuit is outside of scope." Filters remain a Quick Win candidate independent of the POV. |
| Champion(s) | Dan McGrath (process engineer), Sean Halston (ECSM), Todd Davis (lead process engineer, grinding) |
Local problem statement:
The concentrator has 12 AG mills + 24 pebble mills running continuously through a G2 fuzzy logic control layer (SGS). The grind set point is "seat of the pants — we think we're about at the right spot" with no liberation data informing the decision. The classical mineral processing dilemma applies: grind too coarse and you lose recovery; grind too fine and 30-35% of particles drop below 5 microns and act like fluid (bypass losses). Optimizing the grind-recovery curve is valuable but the curve "moves around in 3D." Keith warned that combining grinding + desliming in a first project would make scope "too unwieldy" (Apr 7).
The 42 filters have no individual monitoring — degradation goes undetected for 2-3 days while cascading failures compound. One pilot filter connected to DCS "works really good." Cost to instrument all 42: ~$125K. This is an independent Quick Win.
Bundled initiatives: - TLD-07: AG Mill Throughput Optimization ★★ — ML optimization on top of G2 control, adaptive to ore variability. G2 exists as augmentable foundation. Deferred per Keith (Apr 7) — grinding is the "customer" of desliming, not vice versa. Address after desliming POV proves value. - TLD-22: Filter Performance Monitoring ★★ — instrument 42 filters (~$125K), anomaly detection, same-shift intervention. Quick Win gating the bottleneck. Independent of POV scope. - TLD-11: Concentrator Energy Optimization — reduce overgrinding from pebble mill constraints, high-flux pellet energy demand
Note: TLD-08 (Flotation Recovery) and TLD-32 (Operator Decision Support) moved to TLD-P01 (POV) as of Apr 7 restructuring.
Systems involved: DCS (Foxboro IA), G2 fuzzy logic (SGS), Pi historian, filter instrumentation (to install), autocompo PSI (grind measurement)
Value estimate: $4-9M/yr - AG mill throughput (2-5% improvement across 12 mills) = $2-5M - Filter monitoring = $1-3M (directly gates bottleneck) - Energy savings from reduced overgrinding = $0.5-1M
Confidence: Medium — G2 architecture confirmed, DCS data exists, filter pilot proven. But grinding optimization requires G2/SGS data access (contractual/proprietary risk) and is explicitly deferred.
Implementation approach: 1. Quick Win: Filter instrumentation (~$125K hardware) + anomaly detection. One pilot already works. Scale to all 42. Can proceed independently of POV. 2. G2 augmentation — ML optimization layer on top of existing fuzzy logic for grinding. Deferred until after TLD-P01 POV. 3. Grind set point optimization — link liberation data to grind set point decisions. Requires TLD-P01 results + additional instrumentation.
Dependencies: TLD-P01 (desliming optimization results inform grinding decisions), SGS/G2 data access partnership, filter sensor procurement
Palmer readout alignment: - Scalability: methodology applies to all CLF concentrators / Quick-ROI: filter monitoring is bounded quick win / Palmer named: no — mining-specific
TLD-P03: Pellet Plant Quality & Control¶
| Field | Value |
|---|---|
| Horizon | H1→H2 |
| Corporate project | PRJ-04 — Through-Process Quality & Yield (reframed for mining) |
| Status | identified |
| Champion(s) | Process engineering team (Sean Halston, Todd Davis) |
Local problem statement:
The pelletizing process (14 balling drums → 7 lines → grate-kiln induration at 2,200°F → cooling) is critically dependent on balling operator skill — "the range of finger-painting five-year-old to Bob Ross to Picasso." When skilled operators are unavailable, quality variability spikes. Recirculating loads are the primary diagnostic signal but only detect problems after the cascade has started. Upstream changes (moisture, particle size, material age) affect balling behavior invisibly to operators. Pellet calcium control requires human adjustment every 6 hours and was explicitly called an "easy application test case" by the process engineering team. Clay variability in recent ore has forced running both dryers — a new challenge the plant wasn't designed for.
Bundled initiatives: - TLD-09: Pellet Quality Prediction — predictive model linking concentrate chemistry, flux addition, balling parameters, kiln temperatures to final pellet properties - TLD-34: Pellet Calcium Control Automation ★ — team-designated "easy test case." Human adjustment every 6hr → automated. Data exists. Phase 1 entry point. - TLD-10: Kiln & Grate Temperature Optimization — multi-zone temperature management, energy efficiency
Systems involved: Pellet plant DCS, OCS expert system (pellet side), lab analysis, kiln instrumentation, dryer controls, balling drum controls
Value estimate: $3-9M/yr - Calcium control automation = $0.5-2M - Quality prediction (reduced off-spec + energy savings) = $2-5M - Kiln optimization = $0.5-2M
Confidence: High on calcium control (team-designated quick win, data confirmed). Medium on quality prediction (process well-understood but operator skill variable is hard to model).
Implementation approach: 1. Phase 1 (H1): Calcium control automation — team explicitly called this "easy test case." Well-defined control problem, data exists, low risk. First proof-of-value at Tilden alongside TLD-P10 (HPGR pilot). 2. Phase 2 (H2): Pellet quality prediction — link concentrate moisture/chemistry to kiln zone temperatures and final compressive strength. Real-time adjustment recommendations. 3. Phase 3: Balling operator support — assist less-experienced operators with real-time balling guidance (connected to TLD-P12 Knowledge Capture).
Dependencies: TLD-P01 (ore quality affects pellet quality), lab data access, DCS integration
Palmer readout alignment: - Scalability: methodology applies to all CLF pellet plants / Quick-ROI: calcium control is bounded / Palmer named: no — mining-specific
TLD-P04: Mining Fleet PdM & Lifecycle Intelligence ★★★¶
| Field | Value |
|---|---|
| Horizon | H1→H2→H3 |
| Corporate project | PRJ-03 — Predictive Maintenance Platform |
| Status | validated ★★★ — Day 5 Readout: validated as stepping stone / first win. Tire prediction presented and approved. |
| Champion(s) | Pete Austin (mine maint section mgr, 30 yrs), Chase Lincoln (reliability eng), John Lokowitz (pit electrical) |
Local problem statement:
Tilden operates a fleet of $12M haul trucks (14 fleet) and $27-30M shovels (4 fleet) — 20-25 year assets running 24/7. All fleet maintenance is scheduled on flat operating hours, but not every hour is equal: priority #1 shovel sees 10x more tons than bottom priority, trucks hauling from pit bottom do fundamentally different work than rim trucks. Tire spend is $7.5M/yr (108 tires × $70K). Wheel motors ($300K each) are on their 4th rebuild — uncharted territory where manufacturer recommendations no longer apply. Machine telemetry (fuel burn, duty cycle, component stress) is extracted manually via laptop every 3-4 months. The Bridgestone annual allotment order in August requires accurate forecasting or risks running out (world tire shortage). Pete's team has built impressive Excel-based fleet lifecycle models but they're manual, time-intensive, and person-dependent. Day 5 update: Michelin tire testing also in progress alongside Bridgestone — contractual obligations ending late 2025. AI-driven supplier comparison analytics adds value to tire lifecycle model.
Bundled initiatives: - TLD-02: Heavy Mobile Equipment PdM ★★ — telematics-based PdM for truck engines, drivetrains, wheel motors. Unified fault code ingestion across OEMs (Cat, Komatsu). Fuel burn as engine lifecycle metric. - TLD-19: Tire Management & Prediction ★★★ — per-tire lifecycle model incorporating duty cycle, front/rear position, temperature, pressure, road conditions. Optimized rotation timing and Bridgestone allotment forecasting. Day 5: add Michelin vs. Bridgestone supplier comparison analytics. - TLD-46: Duty-Cycle Based Maintenance ★★ — paradigm shift from hours to actual work done (tonnage, fuel burn, load cycles). Mining industry best practice at Rio Tinto/BHP. Tilden already partially recognizes this (shovel ropes on tonnage intervals). - TLD-47: Fleet Capital Replacement & Lifecycle Planning ★★ — AI-driven multi-year CAPEX model. Optimized replacement sequencing. Repair-vs-replace at each major rebuild decision point. At 120K hours, cumulative component replacement exceeds new truck cost.
Systems involved: Modular dispatch (GPS, cycle times, tonnage), onboard computers (Cat/Komatsu — fuel, duty cycle, fault codes), ELLIPS (CMMS — well-maintained work order history), real-time tire monitoring (temp, pressure, forces), SKF vibration, oil sampling program, Bridgestone supplier portal, vendor rebuild reports
Value estimate: $9-24M/yr - Tire life extension + allotment accuracy = $1-3M (10-15% of $7.5M spend) - Fleet PdM (availability improvement + reduced catastrophic failures) = $3-8M - Duty-cycle optimization (right-sized PM intervals) = $2-5M - Capital lifecycle planning (optimized replacement timing) = $3-8M
Confidence: High — proven at scale in global mining (Rio Tinto, BHP, Vale). Data exists across multiple systems — just fragmented. Pete Austin is the ideal champion: 30 years of domain knowledge, already building the models in Excel, explicitly sees AI opportunity.
Implementation approach: 1. Phase 1 (H1): Tire prediction — richest data, highest stakeholder energy, clear annual cycle (August allotment). Start with front→rear rotation optimization + remaining life prediction. 2. Phase 2 (H1→H2): Duty-cycle maintenance models — start with shovel ropes and truck engines (where the data and physics are clearest). Replace flat hours with composite work metrics (tonnage, fuel burn, load cycles). 3. Phase 3 (H2): Unified fleet PdM — integrate all OEM telemetry through automated data extraction (not every 3-4 months via laptop). Cross-OEM fault code normalization. 4. Phase 4 (H2→H3): Fleet capital lifecycle model — migrate Pete's Excel intelligence into a structured AI system. Multi-year CAPEX plans, repair-vs-replace recommendations, wheel motor rebuild trajectory prediction.
Dependencies: TLD-P09 (Modular↔ELLIPS integration provides data pipeline), OEM data access agreements, onboard data extraction frequency improvement
Palmer readout alignment: - Scalability: CLF mining fleet across all mining operations. Methodology transfers to industrial fleet management at any site. - Quick-ROI: tire prediction has clear annual cycle and dollar anchor ($7.5M/yr) - Palmer named: no specifically — mining-specific. But per-asset optimization on $12-30M assets is intrinsically compelling.
★ The fleet maintenance team is the most data-mature and forward-thinking maintenance group we've encountered at any site. Sophisticated manual systems that need to be digitized and augmented. High per-unit value. Receptive champion.
TLD-P05: Fixed Plant PdM & Failure Analytics¶
| Field | Value |
|---|---|
| Horizon | H1→H2 |
| Corporate project | PRJ-03 + PRJ-01 |
| Status | identified ★★ |
| Champion(s) | George Harmon (reliability eng), Gary (area maintenance) |
Local problem statement:
The concentrator's 12 AG mills and 24 pebble mills, pellet plant kilns (2,200°F), and conveyor systems run continuously in a harsh environment. George Harmon spends hours per week on failure analysis — piecing together breadcrumb trails across ELLIPS (work orders), DCS (continuous monitoring), a 60,000-print drawing database, the relay system, Business Objects, and Power BI. "None of them talking to each other." DCS shows signs 3 months before failures ("current increased, then a leaking seal, then vibration work orders, and eventually the part failed") but this is always assembled post-mortem. Cross-equipment pattern search is nearly impossible in ELLIPS. Gary articulated the most sophisticated deferred maintenance risk framework we've heard: "Pay now or pay later. Paying later is almost always more expensive" — with cause-and-effect years apart and "maintenance amnesia" erasing institutional memory.
Bundled initiatives: - TLD-03: Fixed Plant PdM ★★ — vibration, bearing temperature, motor current analysis for AG mills, kilns, conveyors. DCS breadcrumb trail confirmed as viable signal. - TLD-42: Cross-Asset Failure Pattern Search ★★ — AI-powered search across ELLIPS history + 60K drawings. Natural language queries. Cross-equipment pattern detection. - TLD-41: Deferred Maintenance Risk Quantification ★★ — model the cost trajectory of deferred work. Emotionless risk scoring for mine vs. plant budget allocation.
Systems involved: DCS (Foxboro IA), SKF vibration, ELLIPS (CMMS), Pi historian (1.3B entries), drawing database (60K prints), relay system, Business Objects, Power BI, Oracle (financial)
Value estimate: $4-12M/yr - Fixed plant PdM (mill/kiln/conveyor availability) = $2-5M - Cross-asset failure search (faster analysis + proactive prevention) = $0.5-2M - Deferred maintenance risk (better capital allocation) = $1-5M
Confidence: Medium-High — George's DCS breadcrumb trail description is the clearest articulation of the PdM opportunity at any site. Drawing database and ELLIPS history are well-maintained. Deferred maintenance risk is conceptually powerful but requires significant data integration.
Implementation approach: 1. Quick Win (H1): Cross-asset failure search — RAG over ELLIPS work orders + drawing database. Saves George's team hours/week immediately. 2. Phase 2 (H1→H2): DCS-based PdM — automated cross-system failure pattern detection using Pi historian + ELLIPS. Pre-failure indicator identification. 3. Phase 3 (H2): Deferred maintenance risk model — cost trajectory modeling for major assets. Budget allocation recommendations.
Dependencies: TLD-P09 (data integration), DCS data access, financial system access for cost modeling
Palmer readout alignment: - Scalability: every CLF site has fixed plant assets + CMMS + historians / Quick-ROI: failure search is bounded quick win / Palmer named: no — but PdM is a universal corporate theme
TLD-P06: Drill & Blast Intelligence ★★★¶
| Field | Value |
|---|---|
| Horizon | H2 |
| Corporate project | new — Mining-specific |
| Status | identified ★★★ |
| Champion(s) | Jeff Domann (pit supervisor, 28 yrs — blast crew & yard crew), Tyler Craig (mining engineer) |
Local problem statement:
Blast patterns are currently blanket-loaded — same explosive density in every hole regardless of rock hardness. The drills capture per-hole data (pull-down pressure, rotary speed, GPS, cycle time) that serves as a proxy for hardness. The explosive contractor (Dyno, adjacent to the site) already has trucks capable of auto-loading correct density per hole based on a hardness index. "They have that capability on their trucks now." Both ends of the pipeline exist — the data bridge between drill and explosive truck is the missing piece. ~15,000 holes/year = massive multiplier. Jeff Domann has conceptualized this solution for years but couldn't execute due to data volume: "We've been looking at adjusting our loads, getting energy indexes off the drills... we've never been able to bring that full circle just because there's so much data."
Bundled initiatives: - TLD-05: Drill & Blast Pattern Optimization ★★★ — per-hole explosive density optimization using drill hardness index. Closed-loop: drill data → hardness → Dyno auto-density truck. Saves on soft holes, better fragmentation on hard holes. - TLD-53: Drill Consumable Predictive Ordering — bit life prediction from drill data + formation hardness. Replace Jeff's eyeball inventory with data-driven weekly forecasts.
Systems involved: Drill onboard computers (pull-down pressure, rotary speed, high-precision GPS), Modular dispatch, Dyno explosive delivery software, Vulcan mine model (rock types), ELLIPS (inventory for consumables)
Value estimate: $1-4M/yr - Explosive savings on soft holes + grinding energy reduction from better fragmentation = $1-3M - Consumable forecasting accuracy = $0.2-0.5M
Confidence: High — ★★★. Both technologies exist and are operational. Contractor has the truck capability. Drill data exists per hole. Team has already conceptualized the solution. Gap is purely data integration. Best-articulated mine-specific initiative.
Implementation approach: 1. Phase 1: Extract drill data export format, map to Dyno's input format, build hardness index from pull-down pressure + rotary speed + formation type 2. Phase 2: Pilot on one blast pattern — compare blanket-loaded vs. optimized-loaded fragmentation and grinding performance 3. Phase 3: Full deployment across all patterns. Consumable ordering model as add-on.
Dependencies: Dyno's willingness to integrate (they're adjacent and have the capability), drill data export capability, Vulcan model rock type data access
Palmer readout alignment: - Scalability: all CLF mining operations (Hibbing, United Taconite, Northshore). Drill & blast is universal in open-pit mining. - Quick-ROI: pilot on a single blast pattern is bounded and measurable - Palmer named: no — purely mining-specific. But the "both ends exist, bridge is missing" story is compelling.
★ Jeff Domann is the strongest mine-specific champion — 28 years of domain knowledge, has been thinking about this solution, couldn't execute due to data volume. AI is the enabler he's been waiting for.
TLD-P07: Mine Operations & Dispatch Intelligence ★★★ RYAN'S #2¶
| Field | Value |
|---|---|
| Horizon | H2→H3 |
| Corporate project | PRJ-07 (reframed) + PRJ-02 (reframed) |
| Status | identified ★★★ — Day 5 Readout: Ryan elevated to #2 priority. "Ranks right up there" with feed-forward. Tyler's equipment flow optimization explicitly requested. |
| Champion(s) | Brad Koski (ops mgr), Tyler Craig (mining engineer — readout champion), Molly (dispatch administrator), Andrew Mullen (corporate) |
Local problem statement:
Dispatch is the nerve center of the mine — truck-shovel assignment, crusher routing, and ore blend management all flow through one or two dispatchers managing 14+ trucks and 4 shovels via the Modular system. Non-regular dispatchers take months to become proficient; when the regular dispatcher is on vacation, substitutes "try to get that muscle memory back." Engineering shovel priorities come from Excel, are manually entered into Modular, and errors cascade through tonnage reporting. Real-time plan deviation is a recurring problem — a supervisor recently moved the wrong equipment because he misread the map packet, and the mismatch wasn't caught until the next shift. Andrew Mullen: "If we could have something saying real time, hey, you're getting way off plan." Operator performance scorecards are manual, and 15% overloading damages engines ($1M each) and tires ($70K each).
Bundled initiatives: - TLD-04: Haul Truck Fleet Dispatching Optimization ★★ — AI-assisted dispatch, automated engineering priority ingestion, decision support for non-expert dispatchers, real-time blend optimization - TLD-15: Mine Plan & Production Scheduling ★★ — cascading optimization: annual forecast → production → reagents → shipping → maintenance → what-if scenarios. $1M per mill shutdown. JR's leadership vision. - TLD-50: Real-Time Mine Plan Deviation Alerting ★★ — plan vs. actual comparison engine using Modular dispatch data. Andrew Mullen-endorsed. - TLD-26: Operator Performance & Payload Analytics — automated scorecards from dispatch data, overloading feedback, crew-level performance tracking
Systems involved: Modular dispatch (GPS, cycle times, tonnage), engineering Excel files (shovel priorities), Vulcan (mine plan), daily PowerPoint map packet, crusher DCS, Business Objects
Value estimate: $6-20M/yr - Dispatch optimization (5-15% fleet productivity improvement) = $2-6M - Mine plan scheduling (reduced mill shutdowns, what-if scenarios) = $3-8M - Plan deviation alerting (faster recovery, knowledge accumulation) = $1-3M - Operator analytics (reduced overloading damage) = $1-3M
Confidence: Med-High — Modular data infrastructure exists, dispatcher pain clearly articulated, multiple leaders validated the need. Production scheduling is a longer-term H3 play.
Implementation approach: 1. Phase 1 (H2): Plan deviation alerting — compare daily mine plan priorities vs. actual dispatch. Real-time alerts + deviation documentation. Quick win with immediate visibility. 2. Phase 2 (H2): Operator analytics — automated scorecards from existing dispatch data. Payload/overloading feedback. 3. Phase 3 (H2): Dispatch decision support — AI assistant for non-expert dispatchers. Automated engineering priority ingestion. 4. Phase 4 (H3): Integrated mine planning — full cascading optimization model (JR's vision).
Dependencies: Modular data access, engineering cooperation for priority automation, mine plan format standardization
Palmer readout alignment: - Scalability: dispatch optimization applies to all CLF mining operations / Quick-ROI: plan deviation alerting is bounded / Palmer named: logistics was Palmer's #1 priority at Middletown — mine dispatch is the mining analog
TLD-P08: Mine-to-Dock Logistics Optimization ★★★¶
| Field | Value |
|---|---|
| Horizon | H1→H2 |
| Corporate project | PRJ-07 — Intra-Plant Logistics Optimization (reframed for mining) |
| Status | validated ★★★ — Day 5 Readout: Ryan's #3 priority. "Right up there, top three." |
| Champion(s) | Kevin (train scheduling), Kirk Williams (area mgr transportation), Matt (readout correction — Ryan suggested Matt over Kirk as champion) |
Local problem statement:
Tilden's entire output depends on vessel shipping via a single 130-year-old dock. The vessel schedule is controlled by corporate traffic and changes daily on a 30-day rolling cycle. Three contractors provide 4-6 vessels. Kevin spends 3-4 hours every day replanning train crews. Communication siloes are structural — "there's only certain people who can talk to certain people" — called "intentionally unintelligent" by the team. When vessels don't show, train crews are wasted. When 4 vessels arrive at once, they can't staff enough. The dock has single-source failure points with no ground storage. The railroad team has 950 cars with no lifecycle tracking, and rail breaks occur ~1/week during spring/fall temperature transitions.
Bundled initiatives: - TLD-16: Vessel/Shipping Schedule & Rail Coordination ★★★ — multi-layer logistics optimization. Layer 1: daily schedule optimizer (BCS data + weather + dock status + crew → optimal schedule). Layer 2: predictive disruption model. Layer 3: integrated mine-to-dock flow. - TLD-37: Railroad Asset Maintenance Analytics — Geo car + X-ray car data (years of history) for rail degradation prediction. 950 car fleet lifecycle.
Systems involved: BCS (Best Cargo System — vessel scheduling, years of data), SharePoint (schedule history), Modular (fleet dispatch), ELLIPS (dock/rail maintenance), production model (Excel), weather data, Geo car data, X-ray car data
Value estimate: $2-6M/yr - Crew waste reduction + dock utilization improvement = $2-5M - Railroad maintenance optimization = $0.3-1M
Confidence: High on vessel scheduling — massive existing data in BCS, clear daily pain articulated unprompted as "biggest business challenge this year." Lower on railroad analytics (team self-assessed "data not there yet" in ELLIPS).
Implementation approach: 1. Phase 1 (H1): Daily schedule optimizer — replace Kevin's 3-4hr daily replanning. Takes BCS vessel lineup + weather + dock status + crew availability → generates optimal schedule. Immediate time savings. 2. Phase 2 (H2): Predictive disruption model — learn from BCS historical delays, weather patterns, dock failure modes → probability-weighted schedule with contingencies. 3. Phase 3 (H2): Railroad analytics — aggregate Geo car + X-ray car data, correlate with rail break locations, build degradation models.
Dependencies: Corporate traffic department cooperation for vessel forecast data, BCS data access
Palmer readout alignment: - Scalability: logistics optimization is a universal theme. LS&I rail system serves multiple CLF mining operations. - Quick-ROI: yes — daily schedule optimizer on existing BCS data is bounded and immediately valuable - Palmer named: YES — logistics was Palmer's explicit #1 priority. Coil logistics at Middletown, vessel logistics at Tilden. Same optimization class.
TLD-P09: Ops-Maintenance Data Integration¶
| Field | Value |
|---|---|
| Horizon | H1: Bridge the Gap |
| Corporate project | PRJ-01 — Ops-Maintenance Data Integration |
| Status | identified ★★★ — corporate-validated |
| Champion(s) | Pete Austin (mine maint section mgr), George Harmon (reliability eng), George Beelon (maintenance scheduling), Molly (dispatch admin) |
Local problem statement:
Andrew Mullen (corporate AI program manager) delivered the definitive cross-site validation: "Doesn't matter what CMMS it is... regardless if it's Teams at Middletown, if it's Tabware at Burns Harbor, if it's ELLIPS here, it's not getting done because it's just too cumbersome." At Tilden, the mine-specific manifestation is the Modular Dispatch ↔ ELLIPS disconnect. Dispatch has rich equipment data (operating hours, keys on/off, loaded/unloaded, GPS routes, fuel events) that ELLIPS needs for PM scheduling. Today they're completely disconnected. Equipment hours are manually transcribed from handwritten operator inspection sheets — consuming 4 hours every Monday morning. ELLIPS then makes flawed predictions (pushes PMs out when a machine was down, not running less). Dispatch button-press errors cascade into all downstream reporting — Molly manually corrects 12-hour shifts of data.
Bundled initiatives: - TLD-01: Mining Ops-Maintenance Data Integration ★★★ — unified view of mining operations + maintenance data. Real-time equipment status. Close-the-loop on work orders. Corporate-validated at 3/3 sites. - TLD-45: Modular Dispatch ↔ ELLIPS Automated Integration ★★ — automated operating hours pipeline from dispatch to CMMS. Eliminates 4-hr weekly manual entry. Foundation for duty-cycle maintenance. - TLD-49: Dispatch Status Auto-Correction ★★ — anomaly detection on button-press errors. "Easy thing to spot, but tedious." Replaces Molly's manual shift scanning.
Systems involved: Modular dispatch, ELLIPS (CMMS), DCS (Foxboro IA), drawing database (60K prints), relay system, Business Objects, Power BI, Oracle, Microsoft Fabric (potential data layer)
Value estimate: $4-9M/yr - Modular→ELLIPS integration (eliminated manual entry + accurate PM scheduling) = $1-3M - Dispatch auto-correction (data quality + Molly's time) = $0.3-1M - Broader ops-maint integration (aligned with $2-5M/site estimate from steel sites, mining equipment costs higher) = $2-5M
Confidence: High — pattern validated at 3/3 sites now corporate-confirmed. Modular↔ELLIPS integration is a well-defined solvable problem with clearly identified data on both sides.
Implementation approach: 1. Quick Win: Dispatch status auto-correction — straightforward anomaly detection, training data exists from Molly's corrections 2. Phase 1: Modular → ELLIPS hours pipeline — automated operating hours from dispatch into CMMS meter readings. Eliminates Monday morning manual entry. 3. Phase 2: Enriched data integration — loaded vs unloaded hours, route/grade data, fuel consumption flowing into a data layer. Foundation for TLD-P04 (fleet PdM) and TLD-P04/TLD-46 (duty-cycle maintenance).
Dependencies: IT cooperation, ELLIPS API/import capability, Modular data export capability
Palmer readout alignment: - Scalability: 4/4 sites — universal pattern, Andrew Mullen-confirmed / Quick-ROI: Modular→ELLIPS integration is bounded / Palmer named: no explicitly, but enabling infrastructure for everything else
TLD-P10: HPGR Knowledge Base + PdM Pilot ★★★ LEAD PILOT¶
| Field | Value |
|---|---|
| Horizon | H1: Bridge the Gap |
| Corporate project | PRJ-06 + PRJ-03 — Maintenance Workflow + PdM (combined pilot) |
| Status | validated ★★★ — Day 5 Readout: validated as stepping stone. Ryan: "applicable across many different departments." Knowledge capture pattern confirmed (BLA contract comparison). |
| Champion(s) | Adam Bingham (hybrid maintenance, AI early adopter — already using Copilot), George Harmon (reliability eng) |
Local problem statement:
The HPGR (High Pressure Grinding Rolls) was installed in April 2023. It has 10+ manuals totaling 1,200+ pages that nobody has read. European parts that are unfamiliar: "It's like all European parts. Weird shit." Troubleshooting "always takes a couple days because nobody knows anything about it." The equipment is covered in sensors generating rich data, but nobody's leveraging it. Feed rates dropped unexpectedly in Nov 2025 and the "smoking gun" has not been found despite extensive engineering analysis. The maintenance team collectively nominated the HPGR as the ideal pilot scope — new equipment, clean digital documentation, rich sensors, clear champion.
Bundled initiatives: - TLD-38: HPGR Knowledge Base + PdM Pilot ★★★ — two-for-one pilot: (1) ingest all HPGR manuals/schematics into AI knowledge base for Copilot-based troubleshooting, (2) connect to HPGR sensor data for condition monitoring / anomaly detection. - TLD-33: HPGR Feed Rate Root Cause Analysis ★ — ML investigation of the unexplained feed rate drop. Classic multi-variable problem where human analysis has plateaued.
Systems involved: HPGR OEM sensor system, ELLIPS (work order history building since 2023), Microsoft Copilot, SharePoint (manual storage), DCS, Pi historian
Value estimate: $2-5M/yr - Reduced HPGR downtime + faster troubleshooting = $0.5-2M - Feed rate mystery resolution → sustained high rates = $1-3M - Knowledge preservation for new equipment (template for future purchases) = risk avoidance
Confidence: High — ★★★ team-nominated, documentation is digital and comprehensive, sensors confirmed, champion identified and already using Copilot in the field.
Implementation approach: 1. Week 1-4: Knowledge base — ingest all HPGR manuals, schematics, OEM documentation into searchable AI system. Adam Bingham is the natural user and tester. 2. Week 2-6: Feed rate investigation — ML analysis on DCS data from April 2023 to present. Correlate feed rate with all available process variables. 3. Week 4-12: PdM pilot — connect to HPGR sensor data, build anomaly detection baselines, early warning models. 4. Template creation: Document the approach as a playbook for onboarding future equipment purchases.
Dependencies: HPGR sensor data access (OEM system or DCS?), SharePoint metadata tagging, network connectivity
Palmer readout alignment: - Scalability: the HPGR pilot template applies to any new equipment purchase across CLF. Knowledge base methodology scales to all sites. - Quick-ROI: yes — knowledge base delivers in weeks. Feed rate investigation is bounded. - Palmer named: knowledge capture is Palmer's explicitly stated priority. HPGR pilot is the most concrete, implementation-ready knowledge capture proposal in the Sprint.
★ RECOMMENDATION: This should be the lead pilot project at Tilden. It demonstrates value on both knowledge base and PdM dimensions, has a clear champion, and was nominated by the maintenance team themselves. Success here validates the approach for expansion.
TLD-P11: Maintenance Workflow & Inventory Intelligence¶
| Field | Value |
|---|---|
| Horizon | H1: Bridge the Gap |
| Corporate project | PRJ-06 — Maintenance Workflow Digitization |
| Status | identified ★★ |
| Champion(s) | Adam Bingham (maintenance copilot), warehouse team, Pete Austin (OEM auto-import) |
Local problem statement:
ELLIPS is robust but cumbersome — same pattern as Tabware (Cleveland/Burns Harbor) and SWAMI (Middletown). Adam Bingham is already using Copilot with PDFs for real troubleshooting and Spanish translation. Gary validated the friction: "too cumbersome for people to come back and manually work these systems." The inventory side is worse: ELLIPS search is "terrible," warehouse staff spend 5 minutes to 2 hours per incoming box matching parts. Brad Koski (ops mgr): "Would encompass all of us. Huge win." Every new equipment purchase requires manually creating 5,000+ CLF stock codes that already exist in OEM catalogs — structural root cause of data quality degradation. Parts warehouse has no barcode scanning, all manual.
Bundled initiatives: - TLD-12: Maintenance Workflow Digitization (Copilot) — voice capture, intelligent work order creation, ELLIPS search enhancement. Adam Bingham = champion. - TLD-35: ELLIPS Inventory Master Data Cleanup ★★ — semantic analysis, deduplication, normalized descriptions. Same recipe as MDT-31. Ops side validated: "huge win." - TLD-48: OEM Parts Catalog & PM Procedure Auto-Import ★★ — automated OEM catalog ingestion, cross-reference with existing stock codes, auto-populated PM tasks. Prevents future data degradation. - TLD-30: Parts Warehouse Digitization — barcode/scanner infrastructure for ELLIPS data entry. - TLD-13: Procurement Automation — parts lead time prediction, min/max optimization, automatic PO routing. Parts delays = weekly.
Systems involved: ELLIPS (CMMS/inventory), OEM parts catalogs (Cat, Komatsu), Microsoft Copilot, SharePoint, Oracle (ERP/procurement)
Value estimate: $3-10M/yr - Inventory cleanup + deduplication = $0.5-2M - OEM auto-import (planner time + data quality + equipment readiness) = $0.5-2M - Procurement automation (reduced stockouts + lead time prediction) = $1-3M - Maintenance copilot (workflow efficiency) = $0.5-2M - Warehouse digitization = $0.2-0.5M
Confidence: High on inventory cleanup (proven recipe from Middletown, universal ELLIPS pain confirmed by ops and maintenance). High on OEM auto-import (structured data matching, mature NLP).
Implementation approach: 1. Quick Win: ELLIPS inventory cleanup — same recipe as MDT-31. Semantic dedup, description normalization. Immediate search improvement for everyone. 2. Phase 1: OEM auto-import — automate stock code creation and PM procedure entry for new equipment. Prevents future degradation. 3. Phase 2: Maintenance copilot — Adam Bingham already prototyping. Voice-to-work-order, intelligent search, parts lookup. 4. Phase 3: Procurement intelligence — lead time prediction, min/max optimization, automatic reorder.
Dependencies: ELLIPS data export access, OEM catalog data formats, IT cooperation for barcode infrastructure
Palmer readout alignment: - Scalability: 4/4 sites — every site has the same CMMS friction and inventory pain / Quick-ROI: inventory cleanup delivers in weeks / Palmer named: no explicitly, but maintenance workflow is part of knowledge capture theme
TLD-P12: Mining Knowledge Capture & Virtual SME¶
| Field | Value |
|---|---|
| Horizon | H1→H2 |
| Corporate project | PRJ-09 — Knowledge Capture / Virtual SME (Palmer explicit) |
| Status | identified ★★★ |
| Champion(s) | Adam Bingham (grassroots AI champion), Lynn Casco (mine administrator), Brad Koski (ops mgr), Dan Clarendon (safety/training) |
Local problem statement:
Mining knowledge at Tilden is locked in the heads of people with 25-30 year tenures. Environmental compliance "resides in an outdated spreadsheet or in my head or in two other guys' heads." Night shift 8-month veterans make critical decisions with no reference system. Supervisors are tested by union employees on labor contract questions but nobody's on-site 24/7 to answer. Shift handover emails vary wildly in quality — "hard to figure out exactly what happened." Hundreds of equipment manuals (1,200+ pages for HPGR alone) sit unread. Pi historian has 1.3 billion entries — "if you don't know where to look, you'll never find it." The team asked: "Could this software teach in the moment?"
Bundled initiatives: - TLD-14: Mining Knowledge Capture / Virtual SME ★★★ — searchable AI knowledge base across all mining documentation, Pi historian, equipment manuals. Adam Bingham already prototyping with Copilot. - TLD-43: Maintenance Training Content Generation — AI-generated visual/interactive training from manuals and SOPs. "Being able to show them the procedure instead of reading through a two-page document is going to be huge." - TLD-51: Shift Handover & Ops Knowledge Base ★★ — auto-generate shift summaries from dispatch + supervisor notes. Searchable ops knowledge base. Cross-shift learning. - TLD-52: Labor/BLA Contract Knowledge Assistant ★★ — USW contract chatbot for supervisors. 24/7 availability. Lynn Casco: "It would be amazing." - TLD-27: Environmental Compliance Knowledge System — Brent's compliance knowledge captured and systematized. Regulatory risk reduction.
Systems involved: Pi historian (1.3B entries), equipment manuals, SharePoint, Microsoft Copilot, ELLIPS, shift email templates, USW labor contract, environmental compliance documentation
Value estimate: $2-7M/yr - Knowledge base + Pi searchability = $0.5-2M (reduced troubleshooting time + knowledge preservation) - Shift handover (reduced repeat mistakes + faster recovery) = $0.5-2M - Contract assistant (reduced grievances + supervisor effectiveness) = $0.2-0.5M - Environmental compliance (regulatory risk reduction) = $0.3-1M - Training content (faster onboarding + reduced training incidents) = $0.3-1M
Confidence: High — ★★★ strongest case yet for knowledge capture. Adam Bingham is already building prototypes. Multiple independent knowledge capture needs identified across operations, maintenance, safety, environmental, and HR.
Implementation approach: 1. Quick Win: BLA contract assistant — single well-defined document corpus, proven RAG pattern. Demonstrates AI value to the workforce immediately. Scalable to all USW sites. 2. Phase 1: HPGR knowledge base (via TLD-P10 pilot) — proves the approach on a single equipment, then expands. 3. Phase 2: Shift handover & ops knowledge base — auto-summarize dispatch + supervisor emails into searchable knowledge. Cross-shift learning. 4. Phase 3: Environmental compliance system — capture Brent's knowledge before it walks out the door. 5. Phase 4: Comprehensive Virtual SME — mine-wide knowledge system spanning operations, maintenance, safety, and process.
Dependencies: SharePoint content organization, subject matter expert time for knowledge capture sessions, legal/HR approval for contract chatbot
Palmer readout alignment: - Scalability: 4/4 sites — Andrew Mullen confirmed Burns Harbor has same key-person risk. Virtual SME pattern is cross-site. - Quick-ROI: BLA contract assistant and HPGR knowledge base deliver in weeks - Palmer named: YES — knowledge capture is Palmer's explicitly stated priority. The USW contract chatbot has a unique cross-site story (same union, potentially one model for all USW sites).
TLD-P13: Maintenance Planning & Scheduling¶
| Field | Value |
|---|---|
| Horizon | H1→H2 |
| Corporate project | new |
| Status | identified ★★ |
| Champion(s) | Gary (area maintenance), Pete Austin (mine maint), JR (senior ops/maintenance leader) |
Local problem statement:
Major line repairs cost $1M+ per mill shutdown and require manually extracting data from ELLIPS into Excel, merging into Microsoft Project, and recalculating daily. "Senior supervisors are updating the line repairs every day, trying to get an end date." Crane conflicts, craft availability, shift assignments, contractor coordination, and seasonal road restrictions (mid-March through June — no large parts deliverable) add complexity. Day-to-day maintenance scheduling consumes 4 hours every Monday morning for George Beelon just building the week's plan. Maintenance parts are forecasted with straight-line averages across 65+ item categories.
Bundled initiatives: - TLD-39: Major Repair Schedule Optimization ★★ — AI-assisted scheduling with automated ELLIPS data extraction, critical chain recalculation, resource conflict detection. "$1M per mill shutdown." - TLD-40: Maintenance Resource & Workforce Scheduling — optimized daily crew assignment, absence prediction, overtime optimization within union constraints. - TLD-36: Maintenance Parts & Budget Forecasting — production-correlated parts consumption, what-if scenarios (if 8M tons, how much do wheels/pumps cost?). Migrate Pete's Excel fleet models.
Systems involved: ELLIPS (CMMS), Microsoft Project, resource scheduling, HR/attendance, production model (Excel), Oracle (financial/procurement)
Value estimate: $2-7M/yr - Major repair scheduling (supervisor time + coordination + reduced downtime) = $1-3M - Workforce scheduling (reduced idle time + better utilization) = $0.5-2M - Budget forecasting (accuracy + reduced emergency procurement) = $0.5-2M
Confidence: Med-High — pain clearly articulated with cost anchors ($1M/shutdown). Data exists in ELLIPS. Seasonal road restrictions add a hard planning constraint unique to mining.
Implementation approach: 1. Phase 1: Major repair scheduler — automate ELLIPS → schedule pipeline, real-time updates, resource conflict detection 2. Phase 2: Workforce scheduling — daily crew optimization incorporating absence patterns and union rules 3. Phase 3: Budget forecasting — connect production model to maintenance spend history
Dependencies: ELLIPS data access, MS Project integration, HR data access (privacy considerations)
TLD-P14: Workplace Safety & Inspection Digitization¶
| Field | Value |
|---|---|
| Horizon | H1: Bridge the Gap |
| Corporate project | new (MDT-P07 parallel — safety analytics) |
| Status | identified ★★ |
| Champion(s) | Dan Clarendon (safety/training, 25 years mining + 4 years safety) |
Local problem statement:
MSHA-regulated workplace examinations (Take-5) require every person to complete a quick examination each shift — on paper cards. Supervisors receive 50+ cards per shift and "want to be out in the field." Cards are manually punched into spreadsheets. No reminders for corrective actions. Equipment inspections take 2 hours on paper. The group responded positively to voice/video capture: "Directions to be as simple as taking your smartphone and capturing a video or some pictures." Production reporting follows the same pattern — paper → spreadsheet with multi-day lag.
Bundled initiatives: - TLD-24: Workplace & Equipment Inspection Digitization ★★ — replace 50 paper cards/shift with voice/video capture. Automated corrective action tracking. MSHA compliance dashboard. - TLD-20: Safety Analytics — proximity/fatigue/collision detection leveraging dispatch GPS data. MSHA fine analytics. - TLD-25: Mine Production Reporting Automation — replace paper→spreadsheet pipeline with automated data capture.
Systems involved: Modular dispatch (GPS), MSHA reporting, Take-5 card process, equipment inspection forms, production reporting systems
Value estimate: $1-4M/yr - Inspection digitization (supervisor time + compliance + corrective action tracking) = $0.3-1M - Safety analytics (incident prevention + fine avoidance) = $0.5-2M - Production reporting (lag elimination + data quality) = $0.2-0.5M
Confidence: High on inspection digitization (strong group energy, proven technology). Medium on safety analytics (depends on MSHA data analysis).
Implementation approach: 1. Quick Win: Take-5 digitization — smartphone voice/video capture replacing paper cards. Strongest stakeholder energy of any Day 1 topic. 2. Phase 2: Safety analytics — leverage dispatch GPS for proximity detection, near-miss analytics, MSHA compliance dashboard 3. Phase 3: Production reporting — automated data capture pipeline
Dependencies: Mobile device/connectivity in pit, MSHA data format requirements
Palmer readout alignment: - Scalability: MSHA applies to all mining; OSHA equivalent at steel / Quick-ROI: inspection digitization is bounded / Palmer named: no — but safety is universally important. MDT validated safety analytics with Dave + Palmer + Eric Archer.
TLD-P15: Environmental, Utilities & Geotechnical¶
| Field | Value |
|---|---|
| Horizon | H2 |
| Corporate project | new |
| Status | seed |
| Champion(s) | Brent (environmental manager) |
Local problem statement:
Environmental compliance knowledge lives in 2-3 heads. Selenium contamination is a documented issue (Goose Lake, fish advisories). Utilities/energy forecasting is complex (power contract, ore variability drives energy demand). As the pit deepens, haul road and slope monitoring become more critical.
Bundled initiatives: - TLD-18: Environmental Compliance Analytics — selenium discharge prediction, water treatment optimization - TLD-28: Utilities/Energy Consumption Forecasting — production model → energy demand prediction - TLD-17: Haul Road & Pit Slope Monitoring — geotechnical monitoring, road condition analytics
Value estimate: $1-4M/yr Confidence: Low-Medium — seed status, needs deeper validation Note: Environmental knowledge capture component is in TLD-P12 (Virtual SME). This project covers the analytics and monitoring layer.
TLD-P16: HR & Administrative Operations¶
| Field | Value |
|---|---|
| Horizon | H1 |
| Corporate project | new |
| Status | seed |
| Champion(s) | Lynn Casco (mine administrator) |
Local problem statement:
Overtime prediction via spreadsheets is unreliable ("really hard to predict forward how many hours people work"). Employee onboarding requires multiple manual tickets across ServiceNow, IT, facilities, and ELLIPS. Lynn described 400 people × 2+ weeks vacation entered manually into systems that don't talk to each other. Lower strategic value but high visibility for employee satisfaction.
Bundled initiatives: - TLD-29: HR/Workforce Overtime Forecasting — historical pattern analysis for workforce hours prediction - TLD-44: Employee Onboarding Automation — single-request onboarding triggering all provisioning tasks
Value estimate: $0.3-1M/yr Confidence: Medium — clear pain but low strategic priority relative to production/maintenance
TLD-P17: Mine-to-Concentrator Ore Intelligence (DEFERRED)¶
| Field | Value |
|---|---|
| Horizon | H2: Build the Foundation |
| Corporate project | new — Mining-specific (scalable methodology to all CLF mines) |
| Status | validated ★★★ — Day 5 Readout: Ryan's original #1. DEFERRED: Ryan (Mar 28): "I don't think that's where we want to start. Probably two or three projects down the road." Keith (Apr 7) confirmed: concentrator-first, ore tracking later. Chad (Mar 24) redirected to concentrator. |
| Champion(s) | Todd Davis (lead process engineer, grinding), Dan Collins (process engineer, flotation), + mine engineering TBD (readout feedback: need mine-side champion for ore source data) |
Local problem statement:
Tilden has drill hole data that characterizes ore quality at ~10ft spacing across the entire pit. They also have a concentrator that responds dramatically to ore quality changes — reagent suites, recovery rates, and throughput all shift. But there is no predictive model linking drill data to concentrator response. The current approach is reactive: ore enters the plant, the plant responds, operators adjust, and it takes days to reach steady state. With $50M/yr in reagent spend and ~70% iron recovery, the cost of this reactive gap is enormous.
Tilden has GPS on loading units, position data on trucks, and will "shortly know where the ore tripper is." The data exists to track which blast hole was placed on which line in the crude ore barn. Keith's original vision was to use this data to "understand the variability and with the next step, maybe trying to control the variability to drive stability."
Why deferred: Ryan and Keith concluded (Mar 28, confirmed Apr 7) that combining ore tracking + desliming in one project makes scope "too unwieldy, particularly for a first deployment." Bob Zadel agreed: "put a box around the D-slime process" first, work upstream later. Keith noted ore characterization is "roughly indicative" — bench float tests don't control silica, results are an "indication" not a precise prediction. Starting at the concentrator where data is richer and more reliable de-risks the first project.
Bundled initiatives: - TLD-21: Concentrator Feed-Forward Control ★★★ — ML model linking drill data to optimal concentrator adjustments before ore arrives. The original #1 opportunity — now Phase 2 after desliming POV proves value. - TLD-06: Ore Grade Control & Blend Optimization — integrate geological model, drill data, assay results to optimize daily ore feed blend for target pellet chemistry - TLD-31: Stockpile Ore Distribution Modeling ★★ — model per-section feed quality from existing GPS/truck/tripper data (zero hardware). Team-proposed, hardware exists.
Systems involved: Mine planning (Vulcan 3D model), drill data (pull-down pressure, rotary speed, GPS), Modular dispatch (GPS on every truck, quality data per load), blast hole assay data, stockpile tripper position, crude ore barn feed system
Value estimate: $8-18M/yr - Proactive reagent adjustment from ore prediction = $2.5-5M - Recovery improvement from feed-forward control = $3-8M - Blend optimization for throughput stability = $2.5-5M
Confidence: High — data exists on both sides of the gap (drill data + concentrator DCS), problem clearly articulated by multiple stakeholders, team confirmed "we have the hardware we need" for stockpile modeling. But ore characterization precision is a known limitation (Keith: "roughly indicative").
Implementation approach: 1. Phase 1: Stockpile ore distribution model — Pure data integration, zero hardware. Model per-section feed quality from GPS/truck/tripper data. 2. Phase 2: Feed-forward concentrator model — Link drill hole assay data + geological model to concentrator response. Predict reagent suite adjustments before ore arrives. 3. Phase 3: Full ore-to-concentrator optimization — Closed-loop: drill data → blend optimization → concentrator parameter adjustment → reagent optimization.
Dependencies: TLD-P01 POV results (proves concentrator-side modeling works before adding upstream complexity), drill data format and access, Modular dispatch data export, mine engineering partnership
Palmer readout alignment: - Scalability: methodology applies to Hibbing, United Taconite, Northshore (all variable ore bodies). Principle (upstream data predicts downstream process) applies universally. - Timing: after TLD-P01 POV — "two or three projects down the road" per Ryan
Corporate Project Cross-Reference¶
| Corporate Project | Site Projects | Validation Strength |
|---|---|---|
| PRJ-01: Ops-Maint Integration | TLD-P09, TLD-P05 | ★★★ Strong — corporate-validated by Andrew Mullen across 3 sites. Modular↔ELLIPS disconnect = mine-specific manifestation. Blast quality → maintenance cost dependency. |
| PRJ-02: Production Scheduling | TLD-P07 (reframed) | Partial — mine plan scheduling is H3 strategic play. JR articulated cascading vision. |
| PRJ-03: PdM Platform | TLD-P04, TLD-P05, TLD-P10 | ★★★ Strong — fleet PdM ($12M trucks, $30M shovels), fixed plant PdM (DCS breadcrumb trail confirmed), HPGR pilot (team-nominated). Mining adds mobile fleet dimension. |
| PRJ-04: Quality & Yield | TLD-P02, TLD-P03 (reframed) | ★★ Partial — reframed as concentrator grinding/pellet optimization. Same "through-process quality" principle, mining-specific application. |
| PRJ-05: Cobble & Process Risk | N/A | Does not apply — no hot strip mill. Mining equivalent is blast quality → equipment damage cascade (covered in TLD-P09). |
| PRJ-06: Maint Workflow | TLD-P10, TLD-P11 | ★★★ Strong — HPGR pilot (team-nominated), ELLIPS cleanup (same recipe as MDT-31), OEM auto-import (mining-specific root cause), copilot (champion identified). |
| PRJ-07: Logistics | TLD-P07, TLD-P08 | ★★★ Strong — vessel logistics (★★★ "biggest business challenge"), mine dispatch optimization. Palmer's #1 priority validated in mining context. |
| PRJ-08: Caster Chemistry | N/A | Does not apply — no caster. |
| Knowledge Capture / Virtual SME | TLD-P10, TLD-P12 | ★★★ Strongest — 1,200pg unread HPGR manuals, 1.3B Pi entries, BLA contract chatbot, environmental compliance in 2-3 heads, shift handover gaps. Adam Bingham = grassroots champion already using Copilot. |
New mining-specific projects not in PRJ-01..08: - TLD-P01: Desliming & Recovery Optimization (POV) — ★★★ no steel parallel. $50M reagent anchor. POV confirmed Apr 7. - TLD-P17: Mine-to-Concentrator Ore Intelligence — ★★★ deferred. Bridges ore data to plant response. "Two or three projects down the road." - TLD-P06: Drill & Blast Intelligence — ★★★ both ends of pipeline exist, data bridge missing. Jeff Domann = exceptional champion. - TLD-P13: Maintenance Planning & Scheduling — $1M/mill shutdown, seasonal road constraints (mid-March to June), unique to mining logistics. - TLD-P14: Safety & Inspection Digitization — MSHA-specific. 50 paper cards/shift. Voice capture = strongest quick win energy.
Tilden Roadmap Architecture¶
Tier 1 — Active POV + Signature Projects (highest value, strongest evidence)¶
| Project | Why Tier 1 | Entry Point |
|---|---|---|
| TLD-P01: Desliming & Recovery Optimization ★★★ POV | $50M/yr reagent anchor, ~70% recovery, POV confirmed Apr 7, IE writing charter | 8-week POV: dispersant standardization + live CRP + beaker vision |
| TLD-P04: Mining Fleet PdM & Lifecycle | $12-30M per asset, $7.5M/yr tire spend, champion ready | Tire prediction (August allotment cycle) |
| TLD-P10: HPGR Pilot ★ LEAD | Team-nominated, champion active, best-scoped pilot at any site | Knowledge base in weeks, PdM in months |
| TLD-P08: Mine-to-Dock Logistics | "Biggest business challenge," 3-4hr/day replanning, BCS data | Daily schedule optimizer |
Tier 2 — High-Value Enablers (strong evidence, infrastructure benefit)¶
| Project | Why Tier 2 | Entry Point |
|---|---|---|
| TLD-P09: Ops-Maint Integration | Cross-site validated ×3, corporate-confirmed, enables Tier 1 | Dispatch auto-correction + Modular→ELLIPS pipeline |
| TLD-P11: Maint Workflow & Inventory | Universal ELLIPS pain, proven recipe, enables everything | Inventory cleanup (same recipe as MDT-31) |
| TLD-P06: Drill & Blast | Both pipeline ends exist, champion ready, pure data bridge | Single blast pattern pilot |
| TLD-P12: Knowledge Capture & Virtual SME | Palmer's explicit priority, multiple independent needs, champion active | BLA contract assistant or HPGR KB (via P10) |
Tier 3 — Strategic Plays (longer horizon, higher complexity)¶
| Project | Why Tier 3 | Timing |
|---|---|---|
| TLD-P02: Grinding & Instrumentation | Grinding deferred by Keith (Apr 7). Filter Quick Win can proceed independently. | Filter instrumentation now; grinding after POV |
| TLD-P17: Mine-to-Concentrator Ore Intelligence | Validated at readout but deferred — "two or three projects down the road" (Ryan) | After TLD-P01 POV proves concentrator-side value |
| TLD-P07: Mine Ops & Dispatch | Multiple optimization layers, H2→H3 planning horizon | Plan deviation alerting as entry point (H2) |
| TLD-P03: Pellet Plant Quality | Well-understood process, calcium control quick win | Calcium control automation (H1 entry) |
| TLD-P05: Fixed Plant PdM & Failure Analytics | DCS breadcrumb trail confirmed, deferred risk is conceptually powerful | Cross-asset failure search as quick win |
| TLD-P13: Maintenance Planning & Scheduling | Clear pain, cost anchors, but lower strategic impact | Major repair scheduler |
Tier 4 — Supporting Initiatives (lower priority, log for future)¶
| Project | Status |
|---|---|
| TLD-P14: Safety & Inspection | Quick Win candidate — strong energy but lower $ impact |
| TLD-P15: Environmental, Utilities & Geotechnical | Seed — needs deeper validation |
| TLD-P16: HR & Administrative | Seed — low strategic value |
Site-Specific Notes¶
- Only mine in the engagement — open-pit iron ore mine + concentrator + pellet plant. NOT a steel mill. Many corporate projects (PRJ-01..08) needed reframing.
- Hematite operation using selective flocculation and amine flotation (historically the first mine to produce pellets from both hematite and magnetite ore, 1990-2009; magnetite reserves now exhausted) — ore variability within the hematite body is the root challenge.
- ~990 employees, USW union, 24/7 operations, MSHA-regulated (not OSHA)
- PM rate = 70/30 — significantly better than steel sites (Cleveland 30/70). Gary: "AI telling me to do PMs doesn't help. I already know." Value is NOT more alerts — it's reducing friction so work gets done.
- Most technically open site — no political armor, no defensiveness. Site leader speaks fluently about process chemistry. Testing whether we can match their technical depth.
- Systems landscape: ELLIPS (CMMS — 3rd different one across CLF), Modular (fleet dispatch/GPS), G2/SGS (fuzzy logic grinding control), Foxboro IA DCS (late 1990s), Pi historian (1.3B entries), real-time tire monitoring, SKF vibration, 60K drawing database, BCS (vessel scheduling), Vulcan (mine model)
- Key champions: Keith Holmgren (concentrator, 32 yrs at Cliffs — THE desliming/flotation SME, bus factor risk, POV lead champion), Pete Austin (fleet, 30 yrs), Adam Bingham (grassroots AI, already using Copilot), Jeff Domann (drill & blast, 28 yrs), George Harmon (reliability eng, failure analysis pain), Gary (healthy skeptic, deferred risk framework), Kevin (logistics, daily pain), Lynn Casco (admin, contract assistant)
- Seasonal constraint: Mid-March through June road restrictions — no large parts/equipment deliverable. All planning must account for this window.
- Andrew Mullen cross-site validation: "Doesn't matter what CMMS — it's not getting done because it's too cumbersome." Definitive corporate confirmation of the information flow thesis.