Our team worked with an OEM to understand how the operating environment of some customers was causing the equipment to fail faster than expected. This OEM manufactured Gas Turbines and the components experiencing failure were operating in the hot section. Prior to our engagement machine learning had been applied to high level data available and had not resulted in any useful insights.
Our team therefore started looking at how engineering models of the components could be used to derive better features from high frequency data. This required working extensively with aerospace engineers to encode the models within a pySpark environment. Given the nature of this problem our team worked to test numerous hypotheses against the data. Following a systematic approach of using a Matthews Correlation Coefficient for each hypothesis we were able to work with an initially sceptical hardware engineering team to provide new insights. We successfully managed to identify the insights from which new operating procedures were developed and sent out to the operators of the customers product.
This work demonstrated to our customer that the amount of value residing in datasets is rarely fully utilized and that by applying different thinking and techniques the benefit to the business can be significant.
Please contact to find out more about how we can help you discover and contextualize your existing data to find additional value.