Towards Operation Excellence in Automobile Assembly Analysis Using Hybrid Image Processing

Abstract

The business of a company relies on the delivery of quality products and services with optimal resource usage and it has been a constant endeavor to form timely strategies for improvements in operational efficiency. Many a time product development companies can find opportunities to cut down on repetitive and labour-intensive business processing tasks. Automating the routine process can be a wise business strategy to improve operational efficiency. In this direction, the usage of artificial intelligence concepts like machine learning, deep learning, and reinforcement learning in software product development has been the top choice of many business firms. We believe in, developing value-added differentiators based on recent technological advances in large data management, image processing, and AI/ML algorithm technologies to speed up the drive towards operational excellence. These technologies are at an inflection point, have never been seen before, and definitely can aid in further advancement of business strategies. In this chapter, we discuss one such tool that uses advanced image processing and deep learning algorithms to segment the failure regions from disintegrated automobile assembly parts images. Image segmentation is an aspect of image processing that finds its vast applications in industries and with the advent of machine learning techniques, segmentation has become handier in terms of its computational efficiency. In our technical approach, we use fully convolutional neural networks to segment the region of interest (failure regions) from the image obtained after disintegrating the automobile part, specifically the engine DNox Supply module. One interesting aspect of this work was making segmentation achieve an accuracy of 87% for validation and 98% for training with the sparse dataset. The proposed methodology helped by bringing in intelligent automation instead of manual intensive activity for identifying the region of interest around the failures or abrasions seen in the assembly parts. The generated business reports are shared with the OEM (Original Equipment Manufacturers) for further improvements in the quality of the parts. As illustrated, we can bring in around 40–50% productivity gains along with upwards of 30% cost reduction.