Common Challenges In Manufacturing
100% assurance in cognitive quality control not sampling
-
In high volume production lines, visual inspection becomes the bottleneck process. Conflict between speed and QC
-
Quality control usually done in batches – wasteful loss on batch reject, risk of batch level approvals.
​
​
Calibration of economic and quality levers
-
In high volume production lines, visual inspection becomes the bottleneck process. Conflict between speed and QC
-
Quality control usually done in batches – wasteful loss on batch reject, risk of batchlevel approvals.
Traditional machine vision has not worked
-
Most existing machine vision solutions detect anomalies and not defects.
-
They are not equipped to handle drift in manufacturing process, causing high cost of repurposing
Reliance on individuals - Expensive and shortage of expertise
-
Unavailability of trained Subject Matter Experts (SME). Loss of community knowledge when SME quits
-
Manual Inspection is highly subjective. Results in inconsistent output. Human accuracy rarely exceeds 85%, usually due to employee fatigue. Leads to excess false positives, wasteful rejection
Quality Control In Manufacturing
Traditional QC
-
Manual inspection of basic defects that an eye can see (cognitive and measurable) – at a batch level/QC by sampling
-
Lower accuracy levels, defect leakage and bottlenecks in production line
Evolved QC
-
Using machine vision (cameras) to detect anomalies (anything that doesnt look like a perfect model)
-
Only able to detect defects at a basic level.
-
Image to image comparisons result in high false positives.
Next Gen QC
-
AI powered Automated Cognitive Quality Control Solution
-
Using AI to understand defects and its types and make intelligent decisions that ensure high quality products – just like a good subject matter expert would
-
Fully automated, high accuracy, high speed and consistency.
-
Integrated with Industry 4.0 platforms to enable and effective intervention for better quality
Our Solutions
Vertical solutions for Incoming, In-process(Individual/Mulit-line),
and final Inspection leveraging.
Surface defect - object detection
​Designed to look for specifically defined objects - can be trained and added to defect library
Anomaly detection
​Anomaly Detection aims to identify anomalies - Anything that doesn't look like the original is a defect​
Assembly verification - classification
Designed to look for low number of defects in specific areas​
Print/Object character verification
(OCR/OCV)
Designed to look for Text detection and text recognition
Identification services
Designed to look for anomalies on printed material - labels, barcodes and QR Codes
Industries We Work With
Automotive
-
Fine blanking and forging part - surface defects - dents, damages, scratches, cracks, rust/colour difference
-
Assembly and casting parts - process/part missing, product sorting, counting in packaging and assemblies, Wet leak test
-
Ground and machine parts - Counting in packages and assemblies
-
Radiography - defects on Xray - porosity, shrinkage
-
Print quality - colour, offsets, misprints, smudges. Content error
-
Noise-based inspection (NVH) - Seatbelt, Car horn
-
All Products - product sorting, dimension measurement
Cutting Edge Analytics:
Identifying Process Drifts
Step 1. Training Data to AI
QC Baseline (50 Samples)
Step 2. AI in Production
Raw Image with AI Inference
-
Labels indicates data collected
-
Captures “community knowledge”, which acts as benchmark
-
Matches shop floor information with reality
Step 3. Outcome From Heatmap Production Trend
Month 1 in Production: 514 Instances
Month 2 in Production: 429 Instances
-
Month 1 closely reflecting trained data
-
Month 2 – process drift identified! Major occurrences of defect changed orientation
-
Visual summary of occurrence trend for each type of defect by location
-
Critical input for Preventive Action and continuous improvement
Cutting Edge Analytics:
Recognizing And Clustering Defects
The Cluster Map Shows
-
Pattern in defects will be identified and grouped as clusters
-
Mapping of existing defects
-
Simultaneous mapping of other possible defects/features not trained to AI model
Key Outcomes of the Analysis
-
Helps prioritize focus areas
-
Allows to identify clusters and reclassify them as OK or NG
-
Allows for continuous upskilling for new defect identification and improving accuracy of the end product