Our customer, a global appliance manufacturing company, has faced the problem of repeated and unexpected equipment failures in their production line. The company is known for their commitment to quality and safety standards, while close monitoring of the production process is how the manufacturer achieves this. The manufacturer contacted Intelliarts with the request to find a technological solution for repeated and costly equipment breakdowns, which disrupted their overall plant productivity and caused delays in shipping.
One of the key challenges of this case study was the need to process and analyze the huge datasets of historical and real-time measurements collected from IoT/IIoT sensors. Our solution was to build the scalable data processing unit on top of Dask and deploy it to the AWS Fargate cluster, which allowed us to handle big amounts of data in a scalable, effective manner.
Another problem was the 80% sparsity of data, which meant there were too many gaps in the data collected. This affected the quality of data and the predictive power of the data used in ML models, so we solved this problem by converting the data into a different, more meaningful format.
To increase the predictive power of machine learning algorithms used in our solution and improve the model performance, at the stage of feature engineering, our team of data scientists:
After a series of experiments and proofs of concepts with different ML algorithms, we chose the Extreme Gradient Boosting classifier because of achieving the best score in this case. As a result, our ML-powered solution allowed the manufacturer to predict repeated system failures ahead of time with over 90% accuracy and reduce breakdowns to a minimum acceptable level.
The data science approach allowed us to create the ML model for equipment failure prediction, which eventually helped the customer cut maintenance costs by at least 5%.
The manufacturer got a chance to predict which parts of the equipment were most likely to fail and, thus, maintain or replace those parts just in time and improve their production line performance. As a part of the full-cycle data science project, we also set up the monitoring system for the company and built dashboards so they could track the model results and contact us if there are any significant changes in data or model performance. We also helped with policies and personnel education on how to apply the solution in their day-to-day work.