Industrial datasets and a tool for SISO control loops data visualization and analysis

Abstract

The demand for high performance in industries increases as competitive pressure grows. This opens opportunities for control loop performance monitoring (CPM) techniques, which have been widely proposed over the last three decades. The large number of techniques is valuable, but it brings problems in selecting the right technique for a given application. Industrial data are not easily available due to confidential reasons. Thus, new techniques are commonly tested in a few examples that cover a limited number of scenarios. Also, the tests are not standardized and are not performed using the same dataset. This complicates the comparison of the techniques. This work shares three datasets with different features found on industries, which go from long-range raw data to selected fragments with adequate sampling time. Also, an open-source tool for SISO data visualization and analysis is provided. This tool allows for easy visualization and testing of new techniques.

Introduction

Modern process industries face strong competitive scenarios, which makes operating in high performance an essential key for plant visibility. Aiming to verify whether the controllers, actuators, and sensors are working correctly, control loop performance monitoring (CPM) plays an important role in plant performance (Shardt et al., 2012). CPM techniques have been addressing many topics over the last thirty years, which includes controller performance assessment, non-linearity detection, oscillation detection, and stiction detection (Jelali, 2006).

Its importance makes CPM an always open research area, whose techniques are strongly demanded by the industry. This demand drove the publication of countless works, which is advantageous at the same pace that new problems arise. The first problem is relative to the difficulty to select the right technique for a given application. With a large literature, the selection is an excessively time-consuming task. Another problem is that new researchers are tied to solve overly specific problems or to increase the performance of previously published techniques. This usually induces the development of techniques hard to understand and implement, which are, generally, not well accepted by the industrial community (Bauer et al., 2016).

Since the final application of any CPM technique is the industry, the development of new techniques must be driven by industrial data. Unfortunately, these data are usually confidential and are not shared with the researcher, who are obligated to test their methods to simulated examples or a few datasets freely available. One of these datasets is a benchmark on oscillation detection and diagnosis shared by Jelali and Huang (Jelali and Huang, 2010). Another is a recent repository shared by Bauer et al. (Bauer et al., 2019). Even been freely available and well known by the CPM community, these datasets fail by the fact that they are well behaved and do not include all important features found in real-world applications, which means that good results on these datasets do not guarantee good results in industrial application.

Aiming to better connect industry and academy by solving entirely or partially the problems above mentioned, this work shares new industrial datasets and a tool for SISO data visualization and analysis. The datasets provide examples of different features found in industry and are named:

• SISO-RAW – Raw data collected in two and a half days from an oil and gas company with non-constant sampling time,

• SISO-SEL – Selected fragments from the SISO-RAW dataset, and

• SISO-SAMP – constant sampling time from the SISO-SEL dataset.

Also, the oscillation detection artificial dataset (ODADS) used in previous authors’ work (J.W.V. Dambros et al., 2019) is shared.

The tool, named SISO Viewer, is open source-based and includes fields for data visualization and analysis in time and frequency domains as well as in correlation and parametric plot. With the tool, users can easily load the provided and other datasets and test new methods. The tool was entirely coded by the authors, whose design was based on years of experience studying, applying, comparing, testing, and proposing CPM techniques, mostly on oscillation detection (J.W.V. Dambros et al., 2019; J.W.V. Dambros et al., 2019; J.W.V. Dambros et al., 2019) and diagnosis (Dambros et al., 2016; Dambros et al., 2017; J.W.V. Dambros et al., 2018; J.W.V. Dambros et al., 2018).

Through this work, authors are encouraged to test new methods to as many examples as possible and share their results. The results will be collected in a shared file, where the academic and industrial community can easily compare the performance of the methods, thus making the technique selection easier.

The links to download the datasets, use the tool online, check the shared file with techniques performances, and access the GitHub repository are found and kept updated on www.ufrgs.br/gimscop/repository/sisoviewer.

This work follows with the description of the datasets and the tool in Sections 2 and 3, respectively. A use case on oscillation detection is presented in Section 4. In Section 5, recommendations for the use of the tool are presented and in Section 6 possible cooperation for further improvement of the tool is discussed. The work finalizes with the conclusions in Section 7.