CS-10
ASEAN Automotive Group — AI Predictive Maintenance at Scale
A top-3 ASEAN automotive manufacturer operating 6 assembly plants across Thailand, Indonesia, and Malaysia — producing 480,000 vehicles annually — was experiencing $86M in annual losses from unplanned equipment downtime. Traditional time-based preventive maintenance was both over-maintaining healthy equipment and missing actual failures. Anicalls' AI Predictive Maintenance Agent transformed maintenance operations across all 6 plants — reducing unplanned downtime by 67% and delivering $58M in combined annual saving.
67%Downtime Reduction
$58MAnnual Saving
94%Failure Prediction Accuracy
21 daysAvg Failure Warning Lead Time
Business Challenge
The True Cost of Reactive Maintenance in Automotive Manufacturing
$86M Annual Downtime Loss
Unplanned equipment failures across 6 assembly plants cost the group $86M annually — comprising $52M in direct production loss (vehicles not built during downtime), $18M in emergency maintenance premium costs, $9M in expedited parts sourcing, and $7M in quality rework from suboptimal production conditions during equipment degradation. Assembly line downtime averaged 340 hours per plant per year — 4.3% of available production time.
Time-Based Maintenance Inefficiency
The existing preventive maintenance programme maintained all equipment on fixed intervals — regardless of actual condition. Analysis revealed that 42% of maintenance interventions were performed on equipment that showed no signs of wear (wasted maintenance spend), while 31% of actual failures occurred between maintenance intervals on equipment that degraded faster than expected. The approach cost $28M annually in unnecessary maintenance while still failing to prevent $86M in downtime losses.
No Condition Monitoring Integration
Each plant had legacy SCADA systems generating sensor data — vibration, temperature, pressure, current draw — but none of this data flowed into maintenance decision-making. Sensor data was stored locally, reviewed manually if equipment flagged alarms, and never used for predictive analysis. The maintenance planning team relied on manufacturer recommended intervals and experience — with no analytical tools to detect early degradation signatures in the petabytes of unused sensor data.
Multi-Plant, Multi-OEM Complexity
Six plants across three countries used equipment from 14 different OEMs — Fanuc, KUKA, ABB, Siemens, Mitsubishi, and others — each with proprietary data formats, protocols, and maintenance requirements. No single platform could connect all assets. Maintenance knowledge was fragmented across plant maintenance teams with limited cross-plant learning. A failure mode discovered at the Thailand plant would typically take 18 months to influence maintenance practice in Indonesia.
Solution Delivered
Anicalls AI Predictive Maintenance Agent — 6-Plant ASEAN Deployment
IoT + AI
Universal Asset Condition Monitoring
Anicalls deployed edge computing nodes across all 6 plants — connecting 8,400 equipment assets via OPC-UA, MQTT, and proprietary OEM protocols to a unified data stream. Real-time vibration, temperature, pressure, current draw, and acoustic emission data from all assets flows to the AI platform continuously. The universal data layer eliminated the OEM protocol fragmentation that had previously made cross-asset analysis impossible.
- 8,400 assets connected across 6 plants
- OPC-UA + MQTT + 14 OEM protocol adapters
- Edge computing for real-time local processing
- Sub-10ms sensor data latency
Prediction
AI Failure Prediction Engine
Deep learning anomaly detection models — trained on 4 years of sensor data and 1,240 historical failure events across all 6 plants — predict equipment failures with 94% accuracy and an average 21-day warning lead time. The system identifies the specific failure mode (bearing wear, lubricant degradation, misalignment, electrical fault) and recommends the specific maintenance intervention required, enabling planners to schedule maintenance in planned windows rather than emergency response.
- 94% failure prediction accuracy
- 21-day average failure warning lead time
- Failure mode root cause identification
- Maintenance intervention recommendation
Scheduling
AI Maintenance Schedule Optimisation
AI maintenance scheduling optimises maintenance windows against production schedules — routing predicted maintenance tasks to planned production downtime slots where possible, and calculating the optimal trade-off between intervention timing and production disruption cost when no planned window is available. The scheduler coordinates parts availability, technician skill matching, and OEM specialist scheduling across all 6 plants simultaneously.
- Production-schedule-aware maintenance windows
- Technician skill matching and availability
- OEM specialist coordination automation
- Parts pre-positioning triggered by predictions
AI Workforce Deployment
The Anicalls AI Maintenance Operations Team
AI Asset Health Monitoring Agents
24 AI Asset Health Monitoring Agents — 4 per plant — continuously analyse sensor streams from all 8,400 assets, detecting anomaly signatures and generating health scores for every asset in real time. Each agent monitors 350 assets simultaneously, providing maintenance planners with a live health dashboard that was previously operationally impossible to maintain manually. Alerts are generated in Thai, Bahasa Indonesia, and Malay for local maintenance teams.
AI Maintenance Planning Agents
6 AI Maintenance Planning Agents — one per plant — translate failure predictions into maintenance work orders, coordinate parts procurement, schedule technicians against production windows, and track completion. The agents handle the administrative burden of maintenance planning that previously consumed 60% of human maintenance planner time — freeing human planners to focus on complex diagnostics, OEM relationship management, and maintenance programme improvement.
Cross-Plant Knowledge Agents
Dedicated Knowledge Synthesis Agents identify failure patterns common across multiple plants — enabling maintenance insights discovered in Thailand to immediately inform maintenance planning in Indonesia and Malaysia. The cross-plant learning capability eliminated the 18-month knowledge diffusion lag: new failure patterns are now synthesised and distributed to all plant teams within 48 hours of first detection anywhere in the network.
Technologies Used
The AI Technology Stack Deployed
Platform
Industrial AI Maintenance Platform™
Purpose-built manufacturing AI platform with pre-built connectors for Fanuc, KUKA, ABB, Siemens, Mitsubishi, and 9 additional major OEMs. Deployed as an on-premise edge architecture for real-time processing within each plant, with a centralised cloud analytics layer for cross-plant learning and reporting. Integrated with SAP PM (Plant Maintenance) and IBM Maximo CMMS via pre-built connectors.
- 14 OEM equipment connectors
- Edge + cloud hybrid architecture
- SAP PM and IBM Maximo CMMS integration
- OPC-UA and MQTT protocol support
AI Models
Time-Series Anomaly Detection Models
LSTM and Transformer-based time-series models trained on sensor data from 1,240 historical failure events — learning the characteristic pre-failure signatures for each failure mode across each asset class. Physics-informed neural networks incorporate equipment degradation mechanics (Arrhenius thermal aging, Paris fatigue crack propagation) to improve prediction accuracy beyond pure data-driven approaches, particularly for low-frequency catastrophic failure modes.
- LSTM + Transformer ensemble anomaly detection
- Physics-informed neural network components
- 1,240 historical failure events training set
- Per-asset-class failure mode models
Digital Twin
Assembly Line Digital Twin
Digital twin models of each assembly line — updated in real time from sensor data — enable simulation of maintenance scheduling decisions before implementation. The scheduler uses the digital twin to evaluate the production impact of different maintenance timing options, selecting the scenario that minimises total cost (maintenance intervention cost + production disruption cost + failure risk cost) across the 6-week rolling planning horizon.
- Real-time assembly line digital twin (6 plants)
- Maintenance timing simulation engine
- Total cost optimisation across 6-week horizon
- Constraint-based scheduling (technicians, parts, windows)
Quantified ROI
The Financial Impact at 24 Months
$58M Annual Combined Saving
The $58M annual saving comprised: $42M in production loss prevention (67% downtime reduction × $52M base); $9M reduction in emergency maintenance premium costs; $5M in planned maintenance efficiency gains (42% reduction in unnecessary maintenance interventions); and $2M in parts inventory optimisation (pre-positioned parts procurement replacing emergency sourcing). The $58M saving on a $24M 3-year platform investment delivered a 4.8× ROI in the first 24 months.
12,200 Additional Vehicles Produced
The 67% downtime reduction across 6 plants recovered 1,360 production hours annually — enabling 12,200 additional vehicles to be produced at an average revenue of $18,500 per vehicle. The recovered production capacity was particularly valuable during a regional supply constraint period in 2024-25, enabling the group to fulfil a $226M backorder queue that would otherwise have been cancelled or diverted to competitors.
OEE Improved from 78% to 91%
Overall Equipment Effectiveness (OEE) improved from 78% to 91% — a 13-point gain achieved through the combination of downtime reduction (availability improvement), optimised maintenance scheduling (performance improvement), and predictive intervention before quality-degrading equipment conditions (quality improvement). The 91% OEE placed the group in the top quartile of global automotive OEE benchmarks, ahead of several Japanese and European competitors.
Business Outcomes
Strategic Manufacturing Transformation
Quality
Warranty Claims Reduced 31%
AI-predicted maintenance interventions caught equipment degradation before it caused product quality defects — reducing warranty claims from vehicles built during equipment degradation conditions. Warranty claim rate fell 31% year-on-year, saving an estimated $12M in warranty costs and significantly improving brand quality perception in ASEAN markets. The Thailand plant achieved zero customer warranty claims related to manufacturing defects for the first time in its 22-year history.
Safety
Workplace Safety Improvement
Unplanned equipment failures are a leading cause of workplace injuries in manufacturing — accounting for 64% of all lost-time injuries at the group's plants over the preceding 3 years. The 67% reduction in unplanned failures corresponded to a 58% reduction in maintenance-related lost-time injuries, saving an estimated $4.2M in compensation costs and improving the group's safety rating across all three countries where it operates. The Indonesia plant reached 1,000 days without lost-time injury for the first time.
Carbon
Energy Efficiency Improvement
Degraded equipment consumes more energy than optimal — running inefficiently before failure. AI condition monitoring identified 340 energy-inefficient assets across 6 plants, with targeted maintenance reducing energy consumption by 8.4% in the asset classes addressed. Combined with the elimination of emergency maintenance heating (unplanned overtime, extended hot-work periods), the group reduced manufacturing Scope 1 and 2 emissions by 12,000 tonnes CO₂e annually — contributing to its 2030 carbon intensity targets.
Executive Testimonial
"We had terabytes of sensor data sitting unused for years — we knew it contained value but couldn't access it with our existing tools. Anicalls built the connection layer across 14 different OEM systems and then built the AI on top. Within 6 months, our maintenance teams were receiving 21-day advance warnings of failures that previously hit us without any notice. The $58M saving was transformative, but what I didn't expect was the safety improvement — that has meant more to our teams than any financial metric."
Metrics Dashboard
24-Month Performance Scorecard
67%Unplanned Downtime Reduction
$58MAnnual Saving
94%Failure Prediction Accuracy
21 daysAvg Failure Warning Lead Time
91%OEE (was 78%)
-31%Warranty Claim Rate
-58%Maintenance Safety Incidents
4.8×Platform ROI (24 months)
Predict Failures Before They Happen
See how the Anicalls Industrial AI Maintenance Platform can eliminate unplanned downtime, reduce maintenance costs, and improve OEE across your manufacturing operations.