3.5. Root Cause Analysis

Root cause analysis (RCA) is a technique that performs deep analysis to identify the underlying root causes of failures to avoid their future occurrence. In the literature, we can also find the RCA under synonymous names fault isolation or fault localization. The process is widely used in industry due to the complexity of systems where associating faults with their symptoms is not always trivial. For example, in the case of a seized bearing, the corrective approach is limited to replacing the component with a new one. At the same time, root cause analysis focuses on finding the root cause, e.g., poor lubrication, oil leakage, vibration, etc., which will help avoid future incidents. Classical inference methods rely on deductive, knowledge-based analysis where we can use the following techniques:

A fishbone diagram with various categories of failure or defect cause [97].

Manual analysis by experts is usually of very good quality. Still, it is very time-consuming and requires extensive industry knowledge, which, together with a complex system, limits the applicability of the method. Therefore, it is worth considering the use of data-driven algorithms to automate the process or assist in making it significantly more straightforward. The used techniques inherited features from areas such as Artificial Intelligence, neural networks, graph theory, statistics and automata theory. While analyzing various algorithms used for root cause analysis, Solè et al. [11] categorized them as:

Analysis models may have different properties and performances that affect the purpose of using them. The purpose of the analysis may be to detect root cause based on symptoms, with algorithms capable of detecting, classifying or multi-classifying events occurring simultaneously. The purpose of the analysis may be just to detect or classify root causes based on symptoms or to drill down to explain their nature, origins and related incidents fully. Thanks to this, the algorithms may have the capability to detect, classify or multi-classify events occurring at the same time.