A case study in the province of Modena illustrates how Explainable Artificial Intelligence (XAI) is improving the relationship between machines and workers in the manufacturing sector.
Deep Blue , the leading Italian SME for European research and innovation projects under the Horizon programs (source: European Commission ), together with the European consortium of the XMANAI project coordinated by TXT E-Solutions spa, is committed to improving the way machines and humans interact in the manufacturing industry. Through a “Human-Centered” approach—based on the centrality of the human operator who will collaborate with new intelligent technologies—Deep Blue’s team of experts is committed to making the behavior of Artificial Intelligence (AI) algorithms reliable, understandable, and relevant , promoting trust and effectiveness in the workplace.
Explainable Artificial Intelligence (XAI) aims to eliminate the “black boxes” typical of AI algorithms, allowing operators to fully understand the machines’ decision-making processes. This clarity is essential for increasing confidence in new technologies among those who will use them, especially in sectors like manufacturing, where automated decisions can have a significant impact on safety and productivity .
Making Artificial Intelligence Clear and Transparent
“XAI Explainability” – or Explainable AI – is about the ability to provide the user with understandable, reliable and relevant information at the appropriate level of detail and in an appropriate timeframe about how an artificial intelligence or machine learning application produces its results.
Explainable AI represents a crucial approach in the evolution of Artificial Intelligence and can be implemented in several ways, such as:
- Data visualization : Graphical or visual representations of the data used by the model and the model’s internal processes to show how decisions were made. This may include graphs, charts, or maps that visualize the flow of data and interactions within algorithms.
- Influencing factors : Identify the most influential factors and variables that contributed to a decision, highlighting which inputs had the greatest impact on the results. This allows users to understand which variables were considered by the model and how they influenced the final decisions.
- Model interpretation : Techniques that allow you to assign weights to each input variable and show how these variables influence the model’s predictions. This helps users understand which features are most relevant to the model and how they contribute to its predictions.
- Textual explanations : Natural language or text explanations that describe the model’s decision-making process in a way that humans can understand. This type of explanation is particularly useful because it provides a detailed and easily understandable description of the reasoning behind the AI’s decisions.
The XMANAI project
The XMANAI project, a collaboration of 15 partners from seven European countries, including Italy, Germany, and Spain, has adopted an innovative approach to address this challenge. By combining expertise from research, industry, and academia, the consortium aims to demonstrate the value of explainable AI in solving manufacturing problems and increasing trust in technological innovation.
Within the project, Deep Blue is committed to defining future scenarios involving explainable AI and studying possible interactions between users and AI to maximize explainability, limit human error, and improve overall efficiency. This commitment translates into research and development of tools and methodologies for designing innovative AI user experiences. The project addresses four specific case studies from the manufacturing sector , all with applications in different areas of corporate business:
- Support for preventive maintenance and troubleshooting of numerically controlled machines;
- optimizing the production of an assembly line, to evaluate which factors influence whether or not production objectives are achieved;
- business intelligence support, to forecast demand to support the company’s operational plan;
- support operators in carrying out complex measurement operations and improving their accuracy.
“At XMANAI, we worked on data visualization, testing representations of the outcomes of certain internal steps and the results of AI model analyses. This allowed us to support the understanding of the algorithms’ decision-making process and highlight the inputs that guided the predictions, thus providing a clear explanation of the results obtained and ensuring greater transparency and understanding. This achievement is part of Deep Blue’s vision of developing tools and methodologies for designing innovative AI user experiences, especially in the manufacturing sector,” commented Linda Napoletano, Director and Head of Manufacturing at Deep Blue.
The UXAI tool, developed within the project, allows us to suggest the best data visualizations for making forecasts, creating scenarios, controlling production, or analyzing patterns and trends. Following the tool’s recommendations, for example, interfaces have been designed and tested that allow operators to explore anomalies, make forecasts, and analyze trends in an intuitive and informative way, thus improving the effectiveness of business decisions, as in the case of CNH Industrial in Italy.
The Italian case study: CNH
CNH is a machinery, technology, and services company with over 180 years of history. It operates globally in three segments: agriculture, construction equipment, and financial services. Collaboration with the European partners of the XMANAI project led to the development of a case study at the Modena plant, in collaboration with R&D colleagues at the San Matteo headquarters. This production site currently produces 15,000 transmission trains per year, which are installed on tractors assembled in CNH plants around the world.
The case study focused on machine downtime that interrupts production and maintenance scheduling, a common problem for many manufacturing companies that could be solved with the use of the XMANAI platform . Thanks to the support of XAI technology, operators can diagnose machine errors and efficiently predict the end of component life. This allows for timely replenishment of parts in stock for timely maintenance on machines and optimizes the site’s overall production efficiency, reducing the time and costs associated with unplanned downtime .