Integrating Explainability-by-Design for Transparent and Efficient AI in Manufacturing

Artificial intelligence (AI) is revolutionizing industries, but its complexity often makes it opaque and hard to interpret, and in some cases not accepted by end-users. This opacity has led to the rise of Explainable AI (XAI), which aims to make AI systems’ decisions understandable to humans. However, achieving this transparency is challenging due to the intricate nature of AI models, especially advanced machine learning (ML) methods. The trade-off between accuracy and explainability often complicates this task and efficient XAI strategies need to be considered carefully by decision-makers to maximize the potential of AI.

X-by-Design Paradigm

The X-by-Design approach integrates explainability directly into AI system design, ensuring transparency from the start. This method, validated in the XMANAI project, focuses on embedding explainability as a fundamental requirement in machine learning models. Doing so aims to create systems that inherently offer clear insights into their decision-making processes, avoiding the need for complex post-hoc explanations.

The X-by-Design paradigm in XMANAI involves three key perspectives:

  1. Data Explainability: Understanding the semantics and structure of data to gain insights into input data, monitor potential biases, and ensure data integrity.
  2. Model Explainability: Achieving global and local interpretability of AI models using techniques like surrogate models, feature relevance explanations, and directly interpretable models such as decision trees.
  3. Results Explainability: Providing clear post-hoc explanations through visualizations, text, examples, and counterfactual explanations to help users understand and act on AI outputs.


The XMANAI project also addresses challenges like AI model security and performance, ensuring robust, accurate, and explainable models.

X-by-Design in Action

n addition to the AI/IT services developed in XMANAI to enable X-by-Design in the three perspectives of Data, Model, and Results, the project offers two consultancy-based service axes:

  • X-LEARN Services: For studying manufacturing problems and identifying methodologies to integrate explainability. X-LEARN includes training on XAI concepts, workshops to identify specific needs, and examples from real-world applications to build trust in AI.
  • X-APPLY Services: To integrate XMANAI explainable digital solutions (platform, algorithms, etc.) into manufacturing industrial contexts, providing IT services to enhance XAI adoption in new or existing systems. X-APPLY includes the analysis of the AI status and the implementation of the most effective path to integrate digital solutions based on XAI.
Figure 1: XAI design process

The UXAI tool (https://ai4manufacturing.eu/uxai/) is an important part of X-by-Design consultancy, supporting users in selecting optimal visualizations for explainability, drawing on experiences from XMANAI pilots. It helps design user-friendly XAI interfaces, enhancing the human-centered approach of the X-by-Design process.

X-by-Design was applied in four industrial pilot cases within the XMANAI project:

  1. FORD: Focused on understanding and addressing the specific XAI needs of engine production line operators.
  2. WHIRLPOOL: Tailoring explainability approaches for Central Demand Planners and D2C Marketing & Sales specialists to enhance forecasting and demand analysis.
  3. CNH: Developing user-friendly visualizations for operators and engineers to understand and act on AI model insights.
  4. UNIMETRIK: Providing metrologists and process engineers with visualizations and explanations to optimize measurement scanning processes and improve product quality.

Challenges and Future Perspectives

XAI is essential for making AI decisions transparent and understandable, which is crucial as the impact of AI on human decisions grows. The X-by-Design approach aims to embed explainability into AI systems from the design phase, enhancing performance and human-AI interactions. In manufacturing, this can streamline processes, reduce costs and risks, and ensure business continuity. While challenging, the X-by-Design approach holds significant potential for improving manufacturing productivity and XAI adoption. Please check our website for further information: https://ai4manufacturing.eu