Case Studies

Background

A high-chrome slurry pump on a manganese ore processing circuit was experiencing recurring bearing failures at intervals of 8 to 11 weeks. Each failure resulted in unplanned plant downtime, emergency crane mobilisation, expedited parts procurement and after-hours labour. The mine had no early warning capability — failures were discovered only when equipment tripped or operators reported unusual noise. Bearing replacements were being performed reactively at significant cost and with major disruption to production schedules.

The Data

The mine's SCADA historian had 26 months of continuous data available including bearing temperatures at drive end and non-drive end, motor current on all three phases, discharge pressure, flow rate and seal water consumption. Work order history in the CMMS contained records of 9 bearing failure events over the same period. This data had never been analysed for failure patterns.

The Approach

OreNet extracted and cleaned 26 months of historian data, aligned it with CMMS failure records and engineered features including 4-hour rolling bearing temperature averages, 24-hour temperature rate of change, temperature deviation from process-adjusted baseline, current imbalance percentage and seal water consumption trend. A Random Forest classification model was trained on 20 months of data and validated on the remaining 6 months containing 3 failure events.

The Result

The model achieved a precision of 84% and recall of 89% on the validation period. All 3 bearing failures in the validation set were detected with an average warning lead time of 41 days. The mine was able to schedule all 3 bearing replacements during planned production windows, eliminating emergency callouts and crane hire.

Background

A 6.6kV high-voltage mill motor on the primary mill circuit of a platinum concentrator suffered a catastrophic winding failure resulting in a complete motor burnout. The plant experienced 4 days of lost production. Post-incident investigation revealed the motor had been showing progressive insulation degradation for over 18 months. The warning signs were present in the data. They had never been examined.

The Data

Insulation resistance readings from 6 scheduled shutdown megger tests over 18 months showed a consistent decline from 850 MOhm at commissioning to 120 MOhm in the final reading before failure. Winding temperature RTD data showed a gradual upward trend at constant load over the same period. Motor current data showed increasing phase imbalance developing over the final 6 months. None of this data had been trended or analysed between measurement events.

The Approach

OreNet applied an exponential decay model to the insulation resistance measurement history and a thermal deviation model to the continuous RTD data. The insulation resistance model predicted the motor would reach the critical 50 MOhm threshold approximately 4 months before the actual failure date. The thermal model identified a sustained 9 degree deviation from expected winding temperature beginning 5 months before failure.

The Result

Retrospective analysis confirmed that a planned rewind during the major 6-monthly shutdown 4 months before the failure event would have cost a fraction of the emergency rewind — which, combined with 4 days of lost production, represented a significant financial loss. The insulation resistance trend and winding temperature deviation were both visible in existing data.

Background

A platinum concentrator was replacing mechanical seals on process water pumps reactively — responding to visible leaks and failures as they occurred. Failures were occurring on average every 10 to 14 weeks per pump across a fleet of 8 process water pumps, with each reactive replacement incurring significant cost in parts, labour and production impact.

The Data

Each pump had seal water flow instrumentation feeding the SCADA historian. The historian contained 18 months of continuous seal water flow data at 1-minute intervals. No trending or analysis had been performed on this data. Work order history in the CMMS contained seal failure dates and replacement records for each pump over the same period.

The Approach

OreNet extracted seal water flow data for all 8 pumps and aligned it with CMMS failure records. Feature engineering produced a 7-day rolling slope of seal water consumption measuring the rate of increase over time. Analysis of historical failure events revealed a consistent pattern — seal water consumption began increasing gradually 3 to 5 weeks before each failure event as seal faces wore and required increasing cooling water flow.

The Result

The model achieved an average warning lead time of 18 days across the fleet. Planned seal replacements cost significantly less than reactive replacements, and converting the entire fleet to planned interventions produced substantial annual savings. The data required was already being collected by every pump on the circuit.

Background

A diamond mining operation was experiencing intermittent thermal overload trips on a conveyor drive motor servicing the run-of-mine ore handling system. Each trip caused an unplanned conveyor stoppage disrupting ore delivery to the primary crusher and triggering a production recovery process lasting 2 to 4 hours. The trips were occurring unpredictably during peak production periods and the maintenance team had been unable to identify a root cause despite multiple inspections.

The Data

The mine's MCC historian contained continuous three-phase current data for all conveyor drive motors logged at 30-second intervals. Winding temperature RTD data was available via SCADA. The motor's thermal overload trip history was recorded in the CMMS. Despite all three data sources being available no cross-analysis had been performed.

The Approach

OreNet extracted 4 months of three-phase current data and calculated current imbalance percentage continuously — measuring the deviation between the highest and lowest phase current as a percentage of average current. The NEMA standard for maximum permissible current imbalance is 5%. Analysis revealed phase C was consistently running 11 to 14% above average — well above the NEMA limit. The imbalance was traced to a loose termination at the MCC incomer.

The Result

The root cause was identified from historian data analysis without physical inspection or specialist testing equipment. A 2-hour planned intervention to re-terminate the phase C connection resolved the imbalance completely. A continuous current imbalance monitoring model would have flagged the developing fault 3 weeks before the first thermal trip — allowing the repair during a scheduled maintenance window before any production impact.

Background

A centrifugal pump on the reagent distribution circuit of a platinum concentrator was experiencing accelerated impeller and casing wear at a rate significantly higher than comparable pumps on the same plant. The pump was being pulled for inspection and component replacement far more frequently than its design life suggested. Each intervention required the same parts replacement — impeller, wear rings and casing liners — yet the root cause of the accelerated wear had never been formally identified. The maintenance team attributed the pattern to the abrasive nature of the process fluid and continued replacing components on a reactive basis.

The Data

The historian contained 22 months of continuous data including suction pressure, discharge pressure, flow rate, motor current and pump speed from the variable frequency drive. Process flow demand data from the DCS was also available showing the reagent distribution system setpoint changes across shifts. No analysis had been performed correlating pump operating data with the wear pattern or component replacement frequency.

The Approach

OreNet extracted suction and discharge pressure data alongside flow rate and plotted the pump's actual operating point against its design curve across all operating conditions. Analysis revealed the pump was routinely operating significantly to the left of its best efficiency point during low-demand periods — a classic condition for cavitation development. Suction pressure data showed periodic drops below the calculated net positive suction head required for the pump at those flow conditions. Vibration of broadband character visible in the motor current signature during those periods further confirmed intermittent cavitation. The pattern correlated directly with specific shift handover periods when operators reduced flow setpoints without adjusting pump speed.

The Result

The root cause of the accelerated wear was identified as recurring cavitation driven by operational practice rather than a mechanical or hydraulic design problem. Process adjustments including minimum flow setpoint controls and operator procedural changes eliminated the cavitation conditions. Component replacement frequency reduced substantially in the months following intervention. The historian data contained the complete picture of the problem across the entire 22-month period. The operating practice causing the damage had never been visible to the maintenance team because nobody had looked at where the pump was operating on its curve.

Background

An underground coal mine was operating a dewatering pump station critical to maintaining safe water levels in the working sections. The primary dewatering pump had been in service for 14 months since its last overhaul. No failures had occurred and no alarms had been triggered. From a maintenance perspective the pump was considered healthy. However over the preceding several months the pump had been running for progressively longer periods to achieve the same dewatering duty — operators had been manually compensating by extending run times without identifying or reporting the underlying cause. The change had been gradual enough that no individual shift noticed it as a significant deviation.

The Data

The SCADA historian contained continuous flow rate, discharge pressure, motor current and run hour data going back beyond the last overhaul. The pump control system logged start and stop times automatically. Underground water level sensor data was available from the dewatering sump instrumentation. A complete operational history existed in the historian with no gaps. No performance trending had ever been performed against the pump's post-overhaul baseline.

The Approach

OreNet established a performance baseline for the pump using the first 60 days of data following the last overhaul — a period of known healthy operation. A pump efficiency index was calculated as the ratio of flow delivered to power consumed at comparable operating heads, normalised for any variation in sump level or discharge conditions. This efficiency index was then trended across the full 14-month operating period. The analysis revealed a steady, progressive decline in efficiency beginning approximately 4 months after the overhaul and accelerating over the final 6 months. The decline curve was consistent with progressive impeller wear from the abrasive coal fines present in the dewatering water.

The Result

The efficiency model quantified what operators had been compensating for manually without realising the significance. At the point of analysis the pump was delivering substantially less flow per unit of power than at its post-overhaul baseline — a clear indicator that impeller wear had reached a point where overhaul was required to restore design performance. The pump was pulled for inspection and impeller wear was confirmed at a level consistent with the efficiency trend. The dewatering circuit was returned to design performance following the overhaul. The efficiency decline had been continuously visible in existing historian data for months before the analysis was performed.