The Challenge
A global automotive components manufacturer operating 8 production facilities across North America faced a persistent challenge: unplanned equipment downtime was costing them $35 million annually in lost production, expedited shipping, overtime labor, and emergency repairs.
Their maintenance approach was primarily reactive. Despite having time-based preventive maintenance schedules, critical failures still occurred unexpectedly. Some equipment was over-maintained (wasting resources on components that didn't need replacement), while other equipment failed between scheduled maintenance windows.
With over 500 critical machines including CNC equipment, stamping presses, robotic welders, and assembly automation, the maintenance team was overwhelmed. They needed a way to prioritize their limited resources on equipment that actually needed attention.
The Solution
Jasnova designed and deployed a comprehensive predictive maintenance platform that monitors equipment health in real-time and predicts failures before they occur, giving maintenance teams the advance warning they need to plan repairs efficiently.
Technical Architecture
- IoT Sensor Integration: Deployed 3,000+ sensors across critical equipment measuring vibration, temperature, current draw, acoustic signatures, and process parameters
- Edge Processing: Installed edge computing devices at each facility for local data processing, reducing bandwidth requirements and enabling sub-second anomaly detection
- Time-Series Database: Built on InfluxDB to handle 50 million data points per day with efficient compression and fast queries for historical analysis
- Machine Learning Pipeline: Developed equipment-specific failure prediction models using LSTMs for time-series analysis and gradient boosting for remaining useful life estimation
Prediction Capabilities
The platform provides multiple types of predictions:
- Anomaly Detection: Real-time identification of unusual patterns that may indicate developing problems
- Failure Prediction: 72-hour advance warning of likely failures with confidence scores and recommended actions
- Remaining Useful Life: Estimation of component lifespan to optimize replacement timing
- Root Cause Analysis: Automated identification of factors contributing to equipment degradation
Implementation
Phase 1: Pilot Deployment (Weeks 1-8)
Started with 50 critical machines at a single facility. Installed sensors, established data pipelines, and validated data quality. Collected baseline data to understand normal operating patterns for each equipment type.
Phase 2: Model Development (Weeks 6-16)
Developed failure prediction models using historical maintenance records, sensor data, and domain expertise from maintenance engineers. Created separate models for different failure modes: bearing failures, motor degradation, hydraulic system issues, and electrical faults. Validated models using historical failure events.
Phase 3: Full Rollout (Weeks 12-24)
Expanded deployment to all 8 facilities. Installed additional sensors, deployed edge devices, and trained local maintenance teams. Integrated with the company's CMMS (Computerized Maintenance Management System) for automatic work order generation.
Phase 4: Continuous Improvement (Ongoing)
Implemented feedback loops to continuously improve model accuracy. Maintenance technicians validate predictions, and this feedback is used to retrain models. Added new failure modes as they're identified.
Results
After 18 months of operation, the platform has delivered transformative results:
- 35% Reduction in Unplanned Downtime: From 2,800 hours to 1,820 hours annually across all facilities
- 72-Hour Average Failure Warning: Maintenance teams receive actionable alerts days before failures occur, enabling planned repairs during scheduled downtime
- 25% Reduction in Maintenance Costs: By replacing components at optimal times rather than on fixed schedules or after failure
- 89% Prediction Accuracy: The system correctly identifies 89% of failures with less than 5% false positive rate
- $12M Annual Savings: Combined savings from reduced downtime, optimized maintenance, and avoided emergency repairs
A Real Example
One of the most dramatic successes came when the system detected an anomaly in a critical stamping press. The vibration signature showed a subtle pattern that indicated bearing wear in the main drive system. The prediction model estimated failure within 96 hours with 92% confidence.
The maintenance team scheduled a repair for the upcoming weekend shift. When they opened the bearing housing, they found significant spalling that would have caused catastrophic failure within days. The repair took 6 hours during planned downtime. An unplanned failure of this press would have shut down production for 3+ days, costing over $800,000 in lost output.
"Jasnova's predictive maintenance platform has fundamentally changed how we think about equipment reliability. We've gone from fighting fires to preventing them. Our maintenance team can finally be proactive instead of reactive."
- VP of Manufacturing Operations, Automotive Components Manufacturer
Key Learnings
- Domain Expertise is Essential: The maintenance engineers' knowledge of failure modes and equipment behavior was crucial for developing accurate models
- Data Quality Matters: Significant effort went into ensuring sensor calibration, data validation, and handling of missing data
- Change Management is Critical: Success required buy-in from maintenance technicians who needed to trust and act on the system's predictions
- Start Small, Scale Fast: The pilot phase proved the concept and built credibility, enabling rapid expansion
