Objective:

• Improve vehicle build quality using vision systems, AI & Machine Learning algorithms.

Desired Outcomes:

• Delivery of visually defect-free vehicle to the customer.
• Inspection of fit and finish, painting defects, missing parts and mismatch of parts as per checklist.
• Quantifying in terms of AQI score as per the provided standard.
• Statistical analytics using the obtained data and escalation in case of deviations as defined.
• The system needs to be capable of connecting defects to the stage where the defect is happening and sending alerts/escalations.
• The system should be able to give results immediately after inspection, clearly indicating the zones and places where visual defects need to be corrected.
• The system needs to interact with the ERP system to understand the variant of the vehicle to inspect. 

Current Limitations:

• Skilled men are deployed in the line to inspect the finished vehicle and give dispatch clearance on every shift.
• Inspection needs to be done within the cycle time of <20 seconds on a moving conveyor.
• Inspection is subjective, and perception varies between inspectors and between shifts too.
• Data is manually recorded and not in a form for consumption or analysis.
• Data is entered manually in SAP.
• Occasionally, errors escape to the next stage due to monotony in the inspection.
• More information on existing system constraints:
• Type A and Type B errors – disagreement between man and machine
• Unable to cover all checkpoints (220) in < 20 seconds.
• A vehicle needs to be loaded in a particular orientation to get better images for giving ok and not ok decisions – loading difficulty to an operator to be avoided.
• Difficulty in getting repeatability in results due to allowed variations in fabricated parts.
• The camera cannot clearly differentiate minor dust particles (>1.5 mm) and colour shade differences.
• Some defects will come once in a while, and it is challenging to teach the system frequently.