Dr. Mahmood Tabaddor

Associate Director for Manufacturing Applied Intelligence

Accenture USA

Using modeling and simulation to provide insights into challenging industrial problems from design to manufacturing to regulatory has been my cause for over 25 years. My focus has been on the building of trustworthy models, models whose predictions are demonstrably accurate to support the decision of interest and whose scope of applicability is well understood. The adage that "all models are wrong. Some are useful" is quite often misunderstood. This adage needs to be updated to: all mental models are wrong but some are useful, as a better point of guiding principle to help advance the trustworthiness of models and their predictive power. Verification and Validation (V&V) has been a foundational knowledge base upon which I have built design practices, governance frameworks, and regulatory guidelines enabling the growing use of modeling and simulation technologies. I have tried to contribute the best I can through participation in initiatives such as ASME VVUQ 50 for Advanced Manufacturing, FDA mock submission team for medical device using modeling and simulation using the framework described in ASME V&V 40, and starting the first pilot for a Fire Modeling Working Group at NAFEMS. I started the first use of modeling and simulation to support product Certification through a rigorous and formal V&V process extending the value proposition to now cover product compliance.

Since my interest is in the use of predictive modeling in general, the scope of my work expanded from multi-physics modeling to include machine learning based modeling. I consider the journey and fundamental problem of trustworthiness to be the same for both predictive modeling approaches and that the synergy between the two will only make both more impactful tools in an engineers toolkit. One of the key challenges with machine learning based technologies is that it is not only an engineering tool, it is also a component of a product helping the trend towards autonomy. On the topic of trustworthiness of machine learning based modeling, I started the first ever independent third party assessment program (AI Algorithm Reproducibility Process Verification Mark at UL) in the Testing, Inspection and Certification space. I was also a member of the draft team for the first safety standard for autonomous products, UL 4600. I have been a member of the US TAG for the ISO/IEC JTC 1/SC 42 on Artificial Intelligence, briefly heading the Roadmap Expert Group, and helped develop a series on AI, Data Driven Models and Machine Learning for NAFEMS membership that brings real world engineering practitioners of machine learning modeling technologies.