Project: Find The Root Cause
Rarely has a cooperation started so coincidentally and ended so successfully as the story of our project with a leading company in special vehicle construction.
Challenge of the company:
In the midst of the product lifecycle of its flagship product, the company was faced with a growing number of engine failures. Two years of intensive research could not find the root of the problem.
Coincidence brought IPN and the company’s CTO at the time together at the “Forum Alpbach”. In a riveting discussion about their problem and our approach, we offered to take a closer look.
Then things moved quickly:
The company gave us access to telemetry data from the affected vehicles and put us in touch with the developers of the product. This was followed by a detailed discussion of the faults that occurred and a listing of the hypotheses of the development department. This was followed by data analysis, model building and a final collaborative interpretation of the results and findings.
Project flow:
- After receiving the data, we immediately began an exploratory analysis of the data to understand it and clarify ambiguities as early as possible.
- Various machine learning models were trained to pursue two goals: Identifying the parameters that led to the disturbances from the total of 1,300 measured parameters and predicting disturbances even before they occur.
- In addition to regular consultations with the customer on interim results, the analysis results were jointly interpreted in an on-site customer workshop.
One month until the realization:
After only 25 invested man-days, it was clear that the installation location of the engine in combination with the loading, driving style and a thermally suboptimal cooling air routing led to a steep temperature rise in the engine and the occurring fault messages.
The benefit:
The malfunctions were completely eliminated by a new control of the cooler circuit and subsequent adaptation of the cooler design.
Findings:
Thanks to our pre-selection of relevant parameters from a large set of measurement points (13 out of 1,300 measurement points), the disclosure of temporal correlations and the elimination of “spooky correlations”, we were able to direct the eyes of product developers to the root cause.
This partnership impressively proves that sometimes chance opens the door to groundbreaking success.
