Fault Prediction Algorithm Enhances Industrial Safety and Efficiency

According to a recent study published in the Journal of Chemometrics, a new algorithm can improve fault prediction in industrial processes (1). The new algorithm, called introduces the adaptive slow feature analysis-neighborhood preserving embedding-improved stochastic configuration network (SFA-NPE-ISCN) algorithm, is a tool that is meant to reduce maintenance costs during production processes (1).

In industrial processes, the integration of algorithms and chemometrics has revolutionized quality control, optimization, and efficiency (2). Algorithms, particularly those based on machine learning (ML) and artificial intelligence (AI), analyze vast amounts of process data in real time, enabling predictive maintenance, anomaly detection, and process optimization (3). Chemometrics, the application of mathematical and statistical methods to chemical data, enhances the understanding of complex chemical systems and improves decision-making. By interpreting spectral data and identifying patterns, chemometrics facilitates precise monitoring and control of chemical reactions, ensuring consistent product quality and reducing waste. Together, these technologies enable industries to achieve higher productivity, lower costs, and enhanced product quality, while also contributing to more sustainable and environmentally friendly manufacturing practices (3).

Professional Heavy Industry Engineer/Worker Wearing Safety Uniform and Hard Hat Uses Tablet Computer. Serious Successful Female Industrial Specialist Walking in a Metal Manufacture Warehouse. | Image Credit: © Gorodenkoff - stock.adobe.com

Researchers at Lanzhou University of Technology, led by Xiaoqiang Zhao, explored how machine learning and algorithms could improve safety in industry. As a result, they developed an SFA-NPE-ISCN algorithm designed to improve fault prediction in industrial processes. Industrial processes often operate under varying conditions, making real-time fault detection and accurate fault trend prediction challenging (1). The new SFA-NPE-ISCN algorithm addresses this issue by monitoring fault states in batch processes more effectively (1).

The SFA-NPE-ISCN algorithm conducts slow feature analysis (SFA) to extract the time-varying features of process data. These features are crucial for understanding the dynamic nature of industrial operations and form the basis for updating the Neighborhood Preserving Embedding (NPE) model (1). NPE is used to extract local nearest-neighbor features, reconstructing them with an adaptive update capability (1). This reconstruction relies on square prediction error (SPE) statistics, which serve as indicators of fault states based on the reconstruction error (1).

The research team also made a key modification to their algorithm, employing a hunter-prey optimization (HPO) algorithm to optimize the weights and biases within the stochastic configuration network (SCN), ensuring that the model accurately predicts fault trends (1). The team also integrated singular value decomposition (SVD) and QR decomposition of column rotation to address the ill-posed problems often encountered in SCN, improving the reliability of the improved stochastic configuration network (ISCN) (1).

Once the SPE statistics are formed into a time series, the ISCN model is applied to predict the future states of the process. This comprehensive approach allows for real-time monitoring and precise fault trend prediction, providing a significant advancement over traditional methods (1).

Through case studies of industrial-scale processes, the SFA-NPE-ISCN algorithm and its effectiveness was validated. These case studies demonstrated the algorithm's ability to accurately predict faults and trends, proving its potential to enhance operational safety and efficiency in various industrial settings (1).

By integrating multiple advanced techniques, the research team demonstrated that the SFA-NPE-ISCN algorithm can be used as a tool for industries to maintain stable production and reduce costs associated with unexpected faults and maintenance (1). This development comes at a crucial time as industries worldwide seek to enhance efficiency and reliability amidst increasingly complex operating conditions. The SFA-NPE-ISCN algorithm stands out as a promising solution, offering a path toward safer and more cost-effective industrial operations.

References

(1) Liu, K.; Zhao, X.; Hui, Y.; Jiang, H. An Adaptive Strategy for Time-Varying Batch Process Fault Prediction Based on Stochastic Configuration Network. J. Chemom. 2024, ASAP. DOI: 10.1002/cem.3555

(2) Prando, G. Lowering Dimensions. Nat. Nanotech. 2017, 17, 3113–3118. DOI: 10.1038/nnano.2017.119

(3) Nanotronics, 12 Significant Machine Learning Algorithms in Manufacturing You Need to Know in 2023. Nanotronics. Available at: https://nanotronics.co/thinkspace/machine-learning-algorithms/ (accessed 2024-05-28).