Insights: Root Cause Analysis
Coping with sudden and frequent disruptions, reduced product quality or reduced system efficiency is one of the biggest challenges for our customers. These challenges are often only discovered when they already have a direct impact on operations. The unpleasant thing is that in most cases the causes of these problems are not immediately recognizable.
Due to the impact on operations that has already occurred and the associated costs, it is of the utmost urgency to identify and rectify the causes as quickly as possible. In response, we developed the Root Cause Analysis (RCA) app, which enables us to go from data delivery to detailed root cause analysis in the shortest possible time.
Root cause analysis – background knowledge:
The aim of a root cause analysis is to identify those parameters that explain the difference between normal and impaired operation. If measurement series or data series (e.g. from sensors/actuators) from both operating scenarios are available, suitable machine learning methods can be used to carry out these evaluations in a short time and with the highest accuracy.
It is important to note that machine learning methods to not incorporate plant knowledge or context and therefore only determine statistical correlations that only become meaningful through interpretation and validation by plant experts.
A customer from the special vehicle construction sector commented:
“We didn’t even know where to look in the multitude of measured values and correlations; your approach was like a magnifying glass for us that focused our view on the relevant section of data.”
Root cause analysis – the process:
A key advantage of data-driven root cause analyses is that they follow the same procedure in all cases. We present the basic procedure here:
- Understand:
The be-all and end-all in any data-driven project is an understanding of the issue and the associated data. This enables us to select the relevant data extracts for the analysis and, for example, to analyze the data in detail to eliminate possible “spooky correlations” at an early stage.
- Cleanse:
The initial analysis of the data gives us an overview of the completeness and consistency of the data provided. In this way, questions about the data can be clarified at an early stage and the necessary steps for data preparation can be taken.
- Prepare:
The data is divided into specific data areas and target figures such as quality or efficiency indicators are generated as required.
- Process:
Different machine learning models are applied to the prepared data in order to divide the data into “normal” and “bad” operation. The best models are used to describe the rules for classification.
- Evaluate:
The results of the “Process” step are interpreted and validated together with the system experts.
Root cause analysis with IPN:
In order to provide our customers with the best possible support and to reduce project costs, we have combined the “Prepare”, “Process” and “Evaluate” steps in the RCA app. This enables your system experts to carry out the root cause analysis independently based on the data prepared by us and to arrive at robust results.
- Signal selection:
Definition of the signal to be explained (e.g. efficiency) and the signals that can be used to explain it (e.g. temperatures, speed, etc.) - Data selection:
Selection of the relevant data via the time series display - Accuracy:
Presentation of the model quality for evaluating the analysis results (plausibility) - Ruleset:
Visualization of the set of rules used by the model - Interpret:
Visualization of the raw data used by the model for classification
Your benefit:
With the RCA app, we enable our customers to carry out root cause analyses independently and efficiently. The basic data preparation is carried out by our experts to ensure that the database meets the requirements. By providing an intuitive interface, appealing visualizations and by automating the methodology in the background, the user can concentrate on interpreting and optimizing the findings and carry out all the necessary steps themselves.
Practical application:
Our methodology for root cause analysis has proven itself in the successful handling of various issues, including the rectification of faults in special vehicles and pumps. A concrete example of our newly developed RCA app is the analysis of the turbine of a run-of-river power plant in the hydropower sector. With this app, we enable the operator to independently identify the causes of insufficient turbine power generation in order to take appropriate measures for optimization.
