Session: 10-02 Fan and System Optimisation

Paper Number: 153292

Automated Clustering and the Path Towards Visual Analytics for Enhanced Process Optimization in the Pulp and Paper Industry 

With higher energy prices as well as the desired shift towards sustainability [1], improving the efficiency of their operation is becoming a major challenge of the pulp & paper industry. To achieve this, producers are prioritizing energy savings, optimized machine washing cycles as well as timely fault detection across their plants.

Compressors are one of the key parts of paper production lines. This paper therefore presents an approach to address the aforementioned goals, focusing on compressors and their interaction with other critical components in the plant. To accommodate industry requirements for confidentiality, we explore methods that rely on minimal system knowledge, allowing clients to protect proprietary information.

The scope of this paper is to provide an environment where automatic pattern discovery and outlier detection are interactively editable by users, fostering a collaborative environment for both human-human and human-machine interactions. This approach offers a systematic, algorithm-driven means of capturing patterns and identifying faults, forming the foundation for more complex future algorithms.

We developed and tested a visual analytics [2] pipeline that includes various strategies for window size, overlap size, and projection methods, comparing these results to a fully machine-driven approach that doesn’t include any data visualization technique.

Visual analytics pipelines are capable of managing high-dimensional feature spaces, time-sensitive systems, and hybrid systems that encompass both discrete and continuous signals. These pipelines have demonstrated success across a variety of fields, including medical diagnostics, ornithology, industrial plant monitoring and many others [3, 4].

Both approaches, visual analytics versus the fully machine driven, are discussed in detail, highlighting their advantages and disadvantages. While both leverage dimensionality reduction via an autoencoder as a common starting point, we explore whether further reduction through different techniques are necessary to emphasize behaviors that require attention from monitoring engineers.

This paper applies and tests this strategy in the monitoring of turbomachines and low-instrumentation plants, where operational downtime can be far more costly than the monitoring technology itself. We conclude by presenting an optimized pipeline capable of identifying emergent behaviors and faults, providing timely insights for plant maintenance and energy management without requiring in-depth system knowledge.

References

[1]   Del Rio, Dylan D. Furszyfer, et al. "Decarbonizing the pulp and paper industry: A critical and systematic review of sociotechnical developments and policy options." Renewable and Sustainable Energy Reviews 167 (2022): 112706.

 

[2]   Keim, D., et al. "Mastering the information age: solving problems with Visual Analytics, Eurographics Association, 2010." URL http://www. vismaster. eu/book (2018).

 

[3]   Alsallakh, Bilal, et al. "A Visual Analytics Approach to Segmenting and Labeling Multivariate Time Series Data." EuroVA@ EuroVis. 2014.

 

[4]   Bernard, Jürgen, et al. "VIAL: a unified process for visual interactive labeling." The Visual Computer 34 (2018): 1189-1207.

Presenting Author: David Volponi MAN Energy Solutions Schweiz

Presenting Author Biography: Mechanical Engineer with 5+ years of experience in structural dynamic and static stress analysis, CFD techniques, and Multi-Objective Optimization. Turbomachinery enthusiast, with a problem-solving attitude developed from 7+ years of experience in both industrial and academic innovative environments.

Authors:

David Volponi MAN Energy Solutions Schweiz
Francesco Gant MAN Energy Solutions Schweiz AG
Axel Fiedler MAN Energy Solutions Schweiz
Michael Florian Laubscher MAN Energy Solutions Schweiz AG
Julio José Valero MAN Energy Solutions Schweiz AG