Automating The Inspection Process Of 10,000 Bath Fitting Products (200 Variants) With An Ai-Based Machine Vision

Machine Vision Inspection Of Bath Fitting Products

OVERVIEW

Casting bath fitting products are used to connect different components to control the water flow. They are manufactured with the help of die-cast to mold them into the desired shape. During the procedure of manufacturing, the products go under temperature variations that introduce cracks and other defects (bumps, scratches, depressions, etc) on the surface.

WHY ARE THESE DEFECTS CRITICAL? – Manual inspection of high volume (10,000 per day) with 200 variants is time-consuming and tedious. This leads to many false acceptance and clients has to face bulk complaints from their customers.

Machine Vision Inspection Of Bath Fitting Products

WHAT PROBLEM NEEDS TO BE ADDRESSED?

HOW IS THE PROBLEM BEING ADDRESSED CURRENTLY?

The inspection is completely manual. Once the crack is identified, operators reject the products and reintroduce them to the production line.

WHY AUTOMATION IS NEEDED?

To identify the defects on all the variants and automation would ensure the reduction of the false acceptance rate which is 12-31 percent.

HOW CAN A VISION SYSTEM AND AI SOLVE THIS PROBLEM?

The development of the solution consists of 4 important parts i.e. Image Acquisition, Machine Learning, Solution Deployment, and Accuracy Improvement.

The image acquisition setup includes a camera with red line lights to illuminate the surface of these products with minimum reflection. The setup helps in data collection i.e. acquiring the images of all 200 variants.

Machine Vision Inspection Of Bath Fitting Products

The machine learning part comes after the data collection. It is also called data annotation where images are used to train the AI-based model with the help of annotations (bounding boxes). In this process, all the variants and different defects are trained to the model.

Machine Vision Inspection Of Bath Fitting Products

The trained model is then deployed at the live production line to check for the presence of the cracks in real-time.

In an ideal scenario, not all defects look the same every time. There may be a few defects that are not pre-trained. Therefore, accuracy improvement is required and done by retraining the model until all variations are trained to ensure no unprecedented cracks are missed.

A pretrained library for ‘surface anomaly detection’ is used to identify cracks on the links of the chains

WHAT POTENTIAL BENEFITS CAN BE OBSERVED?

Let’s have a look at the potential benefits to the client after deploying an AI-based vision system –

  • Defects of a minimum 0.5 mm size were identified with more than 96 percent accuracy
  • The average cycle time was reduced from 6-7 seconds to 0.5 seconds
  • Manual intervention was replaced with automation that resulted in consistent progress

Leave a Reply

Schedule A Demo
close slider