This Thai automotive parts manufacturer was a Tier 1 supplier to three major Japanese OEMs, operating two production facilities in the Rayong industrial zone with over 85 CNC machines, stamping presses, and robotic welding cells producing 4.2 million precision components per month. Quality inspection was performed by a team of 38 human inspectors working across three shifts, using a combination of visual inspection, coordinate measuring machines (CMMs), and go/no-go gauges.
Despite rigorous quality protocols, the plant's defect escape rate — defects that reached the customer — was 230 parts per million (ppm), above the OEM-mandated threshold of 150 ppm. Two major customers had issued quality demerits in the previous quarter, and one had indicated that failure to achieve sustained improvement would result in loss of preferred supplier status worth approximately THB 320 million in annual contracts. Human inspector accuracy varied significantly across shifts: day shift detected 91% of defects, while night shift detected only 82%.
The manufacturer had invested THB 15 million in additional CMM equipment the previous year, but the bottleneck was not measurement capacity — it was the speed and consistency of human visual inspection on the production line, where components moved at rates that required inspectors to evaluate one part every 4 seconds.
Pertama Partners initiated the engagement with an AI Readiness Audit that included a detailed study of the manufacturer's quality data, defect taxonomy, and inspection workflows. We analyzed over 45,000 labeled defect images from the quality team's historical records and categorized them across 23 distinct defect types, from surface scratches and porosity to dimensional deviations and coating inconsistencies.
Our AI Pilot Program deployed a computer vision inspection system on two production lines, using high-resolution industrial cameras capturing images at line speed. We developed a custom convolutional neural network optimized for edge deployment, trained on the factory's own defect library and augmented with synthetic defect generation to handle rare defect types with limited training examples. The system was calibrated to flag potential defects with bounding boxes, confidence scores, and defect type classifications displayed on monitors at each inspection station.
The AI Transformation Program then scaled the system to all production lines across both facilities. Team Training was central to the rollout — we positioned the AI as an inspector's assistant that handled the exhausting task of maintaining perfect attention at high line speeds, while human inspectors made the final judgment calls on flagged components. We established a continuous learning pipeline where inspector overrides and new defect types were automatically incorporated into model retraining cycles.
"Pertama Partners understood that in automotive manufacturing, quality is not a feature — it is survival. The AI system they built does not just find defects; it has changed how our entire team thinks about quality."— Somchai Pattanapong, VP Manufacturing
Every transformation starts with a conversation. Let's discuss your challenges and opportunities.
Discuss Your Challenge