AUTHOR=Ji Cheng , Tao Tingting , Wang Jingde , Sun Wei 

TITLE=Multi-Scale Process Monitoring Based on Time-Frequency Analysis and Feature Fusion

JOURNAL=Frontiers in Chemical Engineering

VOLUME=4

YEAR=2022

URL=https://www.frontiersin.org/journals/chemical-engineering/articles/10.3389/fceng.2022.899964

DOI=10.3389/fceng.2022.899964

ISSN=2673-2718

ABSTRACT=<p>Data-driven process monitoring is an important tool to ensure safe production and smooth operation. Generally, implicit information can be mined through data processing and analysis algorithms to detect process disturbances on the basis of historical production data. In industrial practice, signals with different sources of disturbance show different distribution patterns along with the time domain and frequency domain, that is, noise and pulse-type changes are usually contained in the high-frequency portion while most process dynamic is contained in the low-frequency portion. However, feature extraction is usually implemented at a single scale in traditional multivariate statistical algorithms. With this concern, a novel multi-scale process monitoring method is proposed in this work, by which wavelet packet decomposition is first employed for time-frequency analysis. After decomposition, multivariate statistical models are established for each scale to construct process statistics. For the high-frequency part, the classical principal component analysis (PCA) algorithm is adopted to construct squared prediction error (SPE) and Hotelling <inline-formula id="inf1"><mml:math id="m1" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mrow><mml:mo>(</mml:mo><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:math></inline-formula> statistics. While for the low-frequency part, the slow feature analysis (SFA) algorithm is adopted to construct <inline-formula id="inf2"><mml:math id="m2" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>,</mml:mo><mml:mo> </mml:mo><mml:msub><mml:mi>T</mml:mi><mml:mrow><mml:msup><mml:mi>e</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:msub><mml:mo>,</mml:mo><mml:mo> </mml:mo><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo> </mml:mo></mml:mrow></mml:math></inline-formula> and <inline-formula id="inf3"><mml:math id="m3" xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:msub><mml:mi>S</mml:mi><mml:mrow><mml:msup><mml:mi>e</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:msub></mml:mrow></mml:math></inline-formula> statistics for the extraction of the long-term slowly changing trend. Then the monitoring statistics, obtained from each method at different scales, are integrated by a support vector data description (SVDD) method to give a final fault detection decision. The performance of the proposed method is verified on the benchmark Tennessee Eastman Process (TEP) and an industrial continuous catalytic reforming heat exchange unit by comparing with related multivariate statistical methods, which only focus on a single scale.</p>