AI/ML engineering · Oil and gas
Predictive maintenance for pumps, compressors and ESPs
We build predictive maintenance for pumps, compressors and ESPs, including electric submersible pumps and other rotating equipment: condition indicators from vibration, temperature, pressure and electrical data, anomaly detection tuned to each machine, remaining useful life estimates where the data supports them, and work requests raised in your maintenance system. Reliability engineers stay in the loop, and the models and code are yours.
Requirements spec · Oil and gas
Upstream operations system requirements
A requirements specification you can adapt for an upstream production management or surveillance system: master data, acquisition, allocation, deferment, exceptions, reporting and the IT and OT boundary.- 01Eight requirement areas written as testable statements.
- 02Questions on data sources, quality flags and restatement rules.
- 03The IT and OT boundary items security teams ask about.
- 04Seven questions to put to any vendor or development partner.
Free download
Get it on screen in a few seconds.
What the system does, module by module.
- 01
Data acquisition and quality
Vibration, temperature, pressure, flow, motor current and process data from historians, online monitoring systems and handheld routes, with sensor health checks. - 02
Condition indicators
Engineering features such as vibration levels by frequency band, envelope spectra, kurtosis, bearing and gear indicators, and electrical signatures for motors. - 03
Operating context
Normalisation by speed, load and operating mode, so a change in the machine is not confused with a change in the process. - 04
Anomaly detection
Models of normal behaviour for each machine and for peers, flagging deviations early, with the evidence shown to the reliability engineer. - 05
Failure mode recognition
Classification of recognised patterns, such as imbalance, misalignment, bearing defects, cavitation or gas lock, using labelled history where it exists. - 06
Remaining useful life
Degradation trending and survival analysis for components with enough history, with uncertainty shown and used only where it is credible. - 07
ESP and artificial lift analytics
Run-life analysis, motor current, intake pressure and temperature signals, with early warning of trips, wear and operating outside the pump's envelope. - 08
Alert triage and feedback
Alerts ranked by criticality and confidence, with the engineer's verdict feeding back into the models. - 09
Maintenance integration
Work requests, inspections and spare parts triggered in the maintenance system, with the evidence attached and the outcome recorded.
Reference architecture.
Standards and interfaces
What we design against.
- ISO 13373 and ISO 17359
- Condition monitoring and diagnostics of machines: vibration practice and general guidelines for programme design.
- ISO 20816
- Evaluation of machine vibration severity, used for baseline alert limits alongside learned behaviour.
- ISO 13381
- Guidelines for prognostics, which frame how remaining life estimates are produced and reported.
- ISO 14224
- Equipment taxonomy and failure data coding, so failure labels are consistent across assets.
- OPC UA and historian interfaces
- Read-only access to condition and process data.
- Model validation practice
- Back-testing on past failures, false alarm analysis and monitoring in service.
Data model
The core entities.
- Machine and component
- Type, duty, speed, design limits, components and their position in the asset hierarchy.
- Sensor and signal
- Measurement type, location, sample rate, calibration and health.
- Indicator
- Name, window, value, operating state and the signals it was derived from.
- Anomaly alert
- Machine, indicator, deviation, confidence, evidence and status.
- Failure event
- Mode, component, date, repair, downtime and the data around it.
- Life estimate
- Component, method, estimate with interval, assumptions and the model version.
- Verdict
- Engineer's decision on an alert, reason and resulting work.
When custom development is the right call.
Controls designed in, not added later.
Read-only integration
Monitoring and process data is read into the platform. The platform sends no commands to equipment.Engineer decisions
Alerts propose action. Reliability engineers decide, and the decision and reason are recorded.Model governance
Models are versioned and tested, with performance monitored and retraining under change control.Data protection
Equipment and process data is restricted by asset and by vendor agreements.Network boundary
Data crosses a controlled gateway between operational and business networks.
From discovery to a system you run.
- 01
Criticality and data review
We pick the machines and failure modes that matter with reliability and operations, and review the data and failure history available. - 02
Indicators and baselines
Condition indicators and baselines are built, checked against known failures and tuned to reduce false alarms. - 03
Anomaly detection in shadow mode
Detection runs on live data without raising work, and its alerts are reviewed with engineers. - 04
Go live with work integration
Alerts reach engineers and the maintenance system, and verdicts feed back. - 05
Extend to more machines and life estimates
More machine types, ESP analytics and remaining life estimates are added as data allows.
Relevant work
Where to see our work.
Common questions.
- Do we need a history of failures to start?
- Not necessarily. Anomaly detection learns each machine's normal behaviour and flags deviations without labelled failures. Failure mode recognition and remaining life improve as labelled history accumulates.
- How do you keep false alarms down?
- By normalising for operating conditions, comparing a machine with its peers, ranking alerts by confidence and criticality, and tuning thresholds on the engineers' verdicts.
- Does this work for ESPs?
- Yes. ESP analytics use motor current, intake pressure, temperature and vibration where available, to warn of trips and wear, and to track run life across the fleet.
- Does it connect to our maintenance system?
- Yes. Confirmed alerts create work requests with the evidence attached, and the repair outcome returns to improve future detection.
- Who owns the models and code?
- You do. The models, the code and the data stay yours.