Predictive Maintenance Platform

Part of the worlds-biggest-software-project initiative.

An AI-native, open-source platform for sensor data analysis, failure prediction, and maintenance optimisation across industrial assets and fleets.

The Predictive Maintenance Platform ingests multi-sensor data from industrial equipment, detects anomalies, predicts failures, and integrates with CMMS/EAM systems to automate maintenance work. It is built for plant maintenance managers, reliability engineers, and fleet operators who need a vendor-neutral alternative to expensive enterprise SaaS or hardware-locked commercial offerings.

Why Predictive Maintenance Platform?

Incumbent enterprise platforms (IBM Maximo Predict, C3 AI Reliability, GE Vernova APM, Aspen Mtell) are entirely custom-priced with annual contracts in the hundreds of thousands and 6–24 month implementations, locking out SMB and mid-market manufacturers.
Hardware-bundled solutions (TRACTIAN, Augury) create vendor lock-in by requiring proprietary sensors and limiting customers to vibration and temperature data.
No open-source predictive maintenance platforms with AI-native capabilities exist in the commercial-grade segment; open-source CMMS tools (Odoo, CalemEAM, Atlas CMMS) lack predictive ML capabilities.
Most incumbents rely on threshold-based alarms or single-parameter anomaly detection rather than multi-sensor fusion, and few translate sensor patterns into plain-language explanations a technician can act on.
A sensor-agnostic, OPC-UA / MQTT-native architecture removes the hardware lock-in that defines the current market and makes the platform usable on top of existing IIoT investments.

Key Features

Sensor Ingestion and Data Pipeline

Multi-sensor ingestion via OPC-UA and MQTT covering vibration, temperature, current, and pressure
Sensor-agnostic architecture compatible with any IIoT hardware brand
PLC and VFD data ingestion via Modbus TCP and OPC-UA for operating-mode awareness
Historical trend storage with overlay of maintenance events

Anomaly Detection and Health Scoring

Unsupervised anomaly detection (isolation forest, autoencoder, or LSTM-based normal behaviour models)
Asset health score dashboard with fleet-level triage and drill-down views
Multi-sensor fusion combining vibration, temperature, pressure, and current into a single health score
Dynamic threshold adjustment based on operating mode (speed, load) read from PLC/VFD

Prognostics and Root Cause

Remaining Useful Life estimation with confidence intervals
Failure mode classification mapping new anomalies against known failure patterns
LLM-generated plain-language root cause explanations delivered with each alert
Recommended action surfaced alongside detection rather than raw anomaly scores

Alerts, Workflow, and CMMS Integration

Alert management with severity tiering, escalation rules, and routing to email, SMS, and mobile push
Mobile interface for technician alert receipt and acknowledgement
CMMS / EAM work order integration via REST API for SAP PM, IBM Maximo, and SMB CMMS targets
Role-based dashboards for reliability engineers, planners, plant managers, and technicians

Extensibility (Backlog)

Transfer learning framework for bootstrapping models at new sites from cross-fleet data
Spare parts integration linking predicted failure windows to inventory and procurement lead times
Dynamic maintenance scheduling optimiser balancing failure risk, production plan, and technician capacity
Acoustic / ultrasound sensor support with FFT-based bearing defect frequency analysis
OpenAPI-documented developer API and public Python / JavaScript SDKs

AI-Native Advantage

The platform replaces threshold-based alarms with ML models that recognise multi-dimensional failure fingerprints, fusing vibration, temperature, pressure, current, and acoustic data into interpretable Remaining Useful Life estimates. An automated failure mode library learns from historical work order data to classify new anomalies and surface probable cause and recommended action. LLM-generated narratives translate sensor readings into plain-language descriptions ("bearing temperature trending 4°C above baseline over 72 hours, consistent with early-stage lubrication starvation") so technicians can act without data science support. Transfer learning across equipment fleets bootstraps predictions at new sites with minimal retraining data — a capability no surveyed incumbent provides well.

Tech Stack & Deployment

Targeted deployment modes include self-hosted, cloud, and hybrid (mirroring SparkPredict and Aspen Mtell's options) with edge inference for latency-sensitive sites. Open standards underpin the architecture: OPC-UA (IEC 62541) for machine communication, MQTT for sensor telemetry, ISO 13374 for condition monitoring data structures, ISO 13381-1 for prognostics methodology, and MIMOSA / OSA-CBM for maintenance data interoperability. Integration is REST-API-first with documented OpenAPI specs and Python / JavaScript SDKs, alongside connectors for SAP PM, IBM Maximo, and historians such as OSIsoft / AVEVA PI.

Market Context

The global predictive maintenance market is projected at USD 17.11 billion in 2026, growing to USD 97.37 billion by 2034 at a CAGR of 22.2%, with the software segment holding ~55.7% of total share (SNS Insider, Coherent Market Insights). Enterprise platforms (IBM Maximo, C3.ai, GE Predix) are custom-priced with annual contracts in the hundreds of thousands; mid-market tools are subscription-based, and Samsara starts from ~$27/asset/month for fleet use cases. Primary buyers are Plant Maintenance Managers and Reliability Engineers in heavy manufacturing, oil and gas, and utilities; CIOs in asset-intensive industries; and Fleet Operations Directors in logistics and transportation.

Project Status

This project is in the research and specification phase.
Contributions, feedback, and domain expertise are welcome.

Contributing

We welcome contributions from developers, domain experts, and potential users. See CONTRIBUTING.md for guidelines.

Important: All contributions must be your own original work or clearly attributed open-source material with a compatible licence. Copyright infringement and licence violations will not be tolerated and will result in immediate removal of the offending contribution. If you are unsure whether a piece of code, text, or other material is safe to contribute, open an issue and ask before submitting.

Licence

Licence to be determined. See discussion for context.