Image Analytics | Machine Life Expectancy

Business Problem Statement

The maintenance and reliability teams wanted to build an intelligent solution that can help the site engineers to locate the maintenance spots and help in estimating the machine life expectancy. Most of the off-shelf solutions for predictive maintenance offer a computer vision based on the standard industrial datasets that exclude any specific condition that is needed to be considered for maintenance or replacement. Rust and leakage alone doesn’t help in sophisticated setups. It requires a model that can work on 3-D sensors data and predict the life expectancy of the machine.

 

Technical Implementation

  • MS Azure as platform for orchestration
  • Tensorflow and OpenCV for the image processing 
  • Model Training in Azure platform
  • Hosting in Azure using Django framework.

Objectives

Developing a Machine Life Expectancy model with Computer Vision technolog

Solution

AKIRA developed a Machine Learning driven solution to predict maintenance services needs of equipment to improve the life expectancy and deliver value.

Outcome

  • 4X Decrease in TAT for Analytics Reports
  • 74% Increase in ML Model efficiency 
  • Application works asynchronously giving the reliability engineers control over the choice of image that needs to be validated.

Solution Architecture & Workflow

Solution Brief

  • These deep learning solutions are better than the statistical reliability analytics in way that it considers the physical state of the object and does a supervised learning to detect the expectancy whereas the statistical methods map the mean time between the failure and fail to detect the life of the part or machine. They are better for the estimation of life or reliability for the whole lot of products manufactured.
  • The reliability teams at ABB are now leveraging the solution from 1000 image per day analysis to 10K image analysis per day as they have experienced business effectiveness with the Image Processing and Pattern Recognition deployed for the predictive maintenance and reliability engineering.

ML solution hosted on MSFT Azure cloud stack using containerization and wrapper through REST API & Django framework to orchestrate the model output

Machine Learning Model to classify images and identify the type of discharge (i.e. primary driver for the life expectancy of the machine). Further used historical service data and data generated by installed motors & generators to generate maintenance recommendation. Model to predict life expectancy of the machine, which along with services recommendation provide inputs to refine maintenance services strategies

ML model output aggregated and showcased using Power BI to provide visual representation of model output as well as historical process for comparison & efficiency calculations