(222f) Physics-Informed Surrogate Models for Manufacturing Applications

Effective monitoring of a physical or a chemical process requires some form of a representation (inference, clustering, classification) of the measured signals of the process. In model-based monitoring methods, first principles or data-driven models are used to calculate deviations from what is anticipated for the process and what is measured from the available sensors. First-principles models accurately capture the process behavior, but their development can be cumbersome, as physics might be unknown, or the exact relationship or correlation of process inputs and outputs is not accurate or does not exist. Recent advancements in Industry 4.0 technology have led to rich data sets to assist with the development of machine learning data-driven models for process monitoring. These data-driven models can be made more robust and intelligent by incorporating process knowledge (Sansana et al., 2021). The models obtained by fusing process knowledge with data-driven models are novel forms of hybrid modeling. In this context, surrogate models can be seen as a type of hybrid models that are transparent, simple to execute, and wherein domain expert knowledge can be incorporate in the form of intelligent basis functions. In turn, these domain-expert-informed models can be leveraged to related process signals or features with process health monitoring.

In this work, we show the development of a physics-informed surrogate model and its use as an inferential sensor to perform fault classification for process monitoring. Surrogate models are approximate models in the form of explainable algebraic expressions in terms of the inputs of the system. The model outputs are the desired process information or signals that cannot be measured directly, for which it is difficult to have accurate first-principles models. The proposed surrogate models were developed with data of system inputs and outputs for precision machining processes. The process knowledge was incorporated in the form of custom basis functions. We evolved surrogate models using two methodologies: the Automated Learning of Algebraic Models for Optimization (ALAMO) (Cozad et al., 2014; Cozad et al., 2015; Wilson and Sahinidis, 2017); and GPTIPS (Searson, 2015). ALAMO uses machine learning, statistical, and optimization methods to develop a surrogate model, whereas GPTIPS utilizes multigene genetic programming-based symbolic regression to generate models based on a Pareto optimal solution between the model complexity and accuracy. Both methodologies are forms of symbolic regression of the system outputs with respect of its inputs. The models developed were explored for their applicability in classification of faults using K-nearest neighbors.

The process monitoring approach mentioned above was applied to a manufacturing process. Tool wear in machining is the leading cause of tool failure, machine failure, and improper surface finish, resulting in downtime and energy waste in manufacturing. Tool wear depends on machine settings, material and tool properties. Generic surrogate models for tool wear were developed using ALAMO and GPTIPS for data collected from the open literature and in-house run-to-failure tests. Inputs to the surrogate models were the cutting and machine settings, as shown in Fig. 1. Process knowledge was incorporated in the form of custom basis functions, which were formed using literature reports on the mechanisms of tool wear and the key factors contributing to it. These models were used as inferential (soft) sensors for tool wear classification between a new tool and a tool with a flank wear of 0.2 mm, which is commonly a threshold for tool replacement. These soft sensors were shown to improve the classification of tool wear compared to just using the hard sensors of the machine. Further, the surrogate model can be used to improve the prediction of model outputs of the machining model proposed in Awasthi et al. (2021) by providing a generic expression of tool wear propagation. Therefore, utilizing the physics-informed tool wear surrogate model helps increase the effectiveness of the tool wear detection process to avoid downtime and energy waste.

Acknowledgments:

This material is based upon work supported by the U.S. Department of Energy’s Office of Energy Efficiency and Renewable Energy (EERE) under the Advanced Manufacturing Office Award Number DE-EE0007613. We also gratefully acknowledge the Air Force Research Laboratory, Materials and Manufacturing Directorate (AFRL/RXMS) for support via Contract No. FA8650-20-C-5206.

Disclaimer: This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof.”

References

Sansana, J., Joswiak, M. N., Castillo, I., Wang, Z., Rendall, R., Chiang, L. H., & Reis, M. S. (2021). Recent trends on hybrid modeling for Industry 4.0. Computers and Chemical Engineering, 151, 107365. https://doi.org/10.1016/j.compchemeng.2021.107365

Cozad, Alison, Nikolaos V. Sahinidis and David C. Miller. “Learning surrogate models for simulation‐based optimization.” AIChE Journal 60 (2014): 2211-2227.

Cozad, A., Sahinidis, N. V., & Miller, D. C. (2015). A combined first-principles and data-driven approach to model building. Computers and Chemical Engineering, 73, 116–127. https://doi.org/10.1016/j.compchemeng.2014.11.010

Wilson, Z. T., & Sahinidis, N. V. (2017). The ALAMO approach to machine learning. Computers and Chemical Engineering, 106, 785–795. https://doi.org/10.1016/j.compchemeng.2017.02.010

Searson, D. P. (2015). GPTIPS 2: An Open-Source Software Platform for Symbolic DataMining. In A. H. Gandomi, A. H. Alavi, & C. Ryan (Eds.), Handbook of Genetic Programming Applications (pp. 551–573). Cham: Springer International Publishing. https://doi.org/10.1007/978-3-319-20883-1_22

Awasthi, U., Wang, Z., Mannan, N., Pattipati, K., & Bollas, G. M. (2021). Physics-based Modeling and information-theoretic sensor and settings selection for tool wear detection in precision machining. In Process.