How to Develop an AI for Root Cause Analysis in Production
AI-based root cause analysis is a new approach that enables companies to identify the root causes of problems in their production operations and resolve them sustainably. This makes it possible to avoid unplanned downtime and reduce quality defects in manufactured products.
In the article Root Cause Analysis with AI in Production, we have already discussed the opportunities that AI offers for root cause analysis in production and described the four phases of the continuous improvement process.
In this article, we would like to take a closer look at what workflows look like in an AI-based RCA project ("Root Cause Analysis").
Developing AI for Root Cause Analysis
In its white paper AI-Based Root Cause Analysis: Understanding and Optimizing Production Processes, Fraunhofer IAIS outlines workflows that can be used to develop AI for root cause analysis in production. Interdisciplinary collaboration between application experts and data scientists is considered a key success factor for project success, because data scientists alone do not have the process understanding needed to interpret their data results appropriately.
The prototypical workflow of an RCA project looks as follows:
- The application experts define the target event for which causes are to be identified. As a rule, this involves better understanding an error event. In addition, further boundary conditions can be defined under which the target event should be considered.
- For the target event, application experts conduct an initial root cause analysis using classical RCA methods. The result is an initial set of hypotheses about causal relationships and possible causes.
- Data scientists validate the plausibility of the formulated hypotheses using the data. In doing so, they investigate the conditions under which a hypothesis is valid, and what interfering influences lead to deviations from expected behavior. Methodologically, they use so-called interpretable AI models (Explainable AI). Based on their findings, data scientists refine the hypotheses examined and propose new ones. In addition, they prepare their results in a way that is understandable to application experts.
- In a joint review, application experts interpret and validate the AI results and refine their hypotheses. They also discuss with data scientists whether data needs to be corrected or supplemented.
- Steps 3 and 4 are repeated until the project team has identified the most important causes of the target event, and these are simple enough to derive actionable recommendations.
- Application experts formulate the causal relationships and actionable recommendations for daily operations.
In addition to technical complexity, the biggest challenge in an RCA project is ensuring productive collaboration between application experts and data scientists. Establishing a clear understanding of roles and moderating different perspectives is particularly important to create a creative and solution-oriented work environment.
Lack of Expertise as a Barrier to AI Innovation
A major obstacle for manufacturing companies is the shortage of data scientists (see State of AI in the Enterprise - 3rd Edition (Deloitte) and a general lack of expertise in managing AI projects. In the article AI Innovation: What Skills Do You Actually Need? we have compiled the necessary competencies for you.
To date, only large companies can afford to invest in their own digitalization departments and data scientists to implement data projects completely in-house. For everyone else, collaboration with a suitable external provider is the more economical option.
An interesting perspective for the next 5 years is that AI assistant systems will emerge in the production environment that automate the role of the data scientist for AI-based root cause analysis. Realizing such solutions requires new concepts for human-AI interaction that enable non-AI experts to interpret AI results and derive actions from them. This topic is part of active research that we are involved in as part of the GeMeKI project.
Conclusion and Outlook
In our assessment, manufacturing companies in Germany will benefit significantly from AI-based root cause analysis in the coming years to maintain and expand their competitiveness. Whether to sustainably address problems in their own operations or to ensure the smooth deployment of their products in customer applications.
Our daily work at aiXbrain is to make these capabilities quickly and easily accessible, for example with aiXbrain Dataray, an AI software solution for rapid diagnosis of process workflows. We also actively participate in the GeMeKI research project in the human-centered design of such AI-based assistant tools.
