A Review on Data-Driven Process Monitoring Methods: Characterization and Mining of Industrial Data

Round 1

Reviewer 1 Report

In this work, the author reviewed the Data-driven Process Monitoring Methods. The topic is very interesting due to the emergence of new data-driven techniques and improved computer performance for process monitoring. Furthermore, the general structure of the article is excellent, and the authors did their best to incorporate all details of process monitoring as possible in this review. As a result, this paper will be an excellent source of information for the researchers working in this field.    

Few cosmetic remarks:

1- Can the "industry data" in the title change to industrial data? Also, line 15

2- line 11 and 63, intentions or attentions?

3- line 43, mechanism models or mechanistic models?

4- line 108, "proposed to extraction the", to extract

5- line 135, "characteristics, which could be defined as that at least one", needs to be modified

6- line 353, "considered by hid-den state", hidden

7- line 353, "proposed moving PCA and", moving window PCA?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Because it is difficult for operators to detect abnormal process deviations and make proper decisions to eliminate them at their early stage due to the increasing scale of chemical production and equipment complexity. P. 1

Industrial processes are not limited to a single operating condition, and characteristics of normal operating conditions also vary with different processes, which brings great challenges to the monitoring of industrial processes. P. 3

With these sentences, the authors indicate that it is becoming increasingly difficult for operators (people) to make timely and correct decisions in a current complex situation that consists of various interrelated chemical processes that operate on an increasingly larger scale. This starting point is not reflected in the assessment of the various options.

Therefore, data in actual production often display multimode characteristics, which could be defined as that at least one variable doesn’t follow a single steady operating condition due to various changes in production loads, feed flow and set points. However, traditional statistic process monitoring models are established under the assumption that the process is operated at a single  stable working point. P. 5

The time-varying characteristics violate the assumption of traditional multivariate statistic process monitoring that the process is time-independent, and therefore limit the application of industrial process monitoring. P. 5

Considering that causal logic among process variables can also be described as time delayed correlation between each pair of variables, several correlation analysis methods have been introduced to identify the fault propagation p. 23

Time therefore plays a crucial role in determining the process variables. Current statistical procedures are not (yet) sufficient as stated by the authors. Nevertheless, the relationship between the operator (human) and these limitations of the machine (AI) is not discussed. The question of whether and how the causal relationships within the process can be explained is only answered from the possibility of even better calculations by the machine (AI) but not from the perspective that it will be even more complicated for the operator to understand and diagnose and make the right decisions based on that. This means that these new possibilities, as stated here, can lead to more errors rather than fewer.

If causal reasoning among these variables could be obtained, the fault propagation could be displayed in a network diagram and then the real root cause of the fault could be located. The acquisition of causal reasoning models among process variables has always been a challenge p. 8

Moreover, the response of control systems when a fault occurs will result in the changes in variable correlation, making the diagnosis results inconsistent with practical situations p. 27

How to effectively employ real time data to capture the causal logic among process variables or adaptively update the causal network established using historical data from normal operating conditions can be further investigated for the identification of fault propagation path. P.27

Although this could be a promising approach, it has been the least elaborated. Also, the question is not answered when even more connections are included (for the causal reasoning by the machine (AI)) what this means for the interpretation possibilities of the results by the operator (situational awareness) of the current situation.

Although this is a good study into the current situation of data-driven process monitoring, it is in my opinion insufficiently elaborated in the relationship (operator) and machine (AI). In my opinion, this means that the challenges and prospects as well as the conclusions are too brief and insufficiently elaborated. It is recommended that the question of the why (AI) and the what (reducing errors by operators) be addressed with more emphasis.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The paper proposes a review of different data-driven processing monitoring methods, focusing mostly on chemical processes.

After a brief introduction to fault detection and diagnosis, and to some of the required terminology such as distributed control systems, principal component analysis, etc., a general data-driven process monitoring framework is presented.
The section, in particular, introduces the different components of a framework for monitoring industrial processes, such as the definition of normal operating conditions, the data processing approaches, the selection of feature extraction methods and the monitoring and diagnosis components, their relationships and some approaches typically used to address the specific problems.
The actual review follows, introducing the approaches used in the literature for the different components of the previously introduced framework.
Some challenges not yet resolved (or not effectively resolved) are then listed, proposing some possible solutions.
Some conclusions close the paper.

The paper is overall well written and easy to follow.
The review, to the best of my knowledge, seems sufficiently exhaustive, covering all the problems related to process monitoring.

A comment I would like to make to the authors is to introduce some comparative/summary tables to break up the document and make it less "heavy", especially with regard to section 2.
The tables in section 3 already go in this direction.
Perhaps a table with the different approaches followed in the literature to solve the different issues could make the document more readable. It could also help the reader to easily find, among the cited literature, the components of monitoring in which he/she is interested most.
The introduction of some additional figures could also benefit the document. In order to better group and characterize the different approaches, providing an overview of the problems faced.

Continuous process monitoring approaches are divided into two big categories: multivariate statistics and novel machine learning algorithms.
However, this second category also includes approaches based on expert systems, which typically have nothing to do with machine learning.
I would suggest to the authors to check this aspect in section 3.

Still in section 3, it is said that one of the reasons for the scarce use of random forests is the fact that these approaches are considered as a black box.
However, most machine learning-based approaches are often black boxes.
In this regard, it might be appropriate for the authors to discuss any problems related to the explanation of the models (in the context of the explanable AI) applied to process monitoring.
Explainable AI, in particular, could be used to explain models and thus make them more trustworthy for users who use them.
Perhaps this topic could be covered in section 4?

Why does the document focus on the chemical industry? The approaches described seem general enough to be applied to many industries, not necessarily chemical ones.

Despite the previous comments, the paper is, in my opinion, well done, sufficiently self-contained, and well structured.
I have only a few minor comments:
 - line 39, that "take advantage" sounds weird in English. Maybe it should be a "Taking advantage"? Or "In order to take advantage"?
 - line 75, "process" should probably be "processes"
 - line 104, "process" should probably be "processes"
 - line 207, "aim" should probably be "aims"
 - line 204, "method" should probably be "methods"
 - line 224, please, check the sentence "Feature extraction is to obtain". It should probably be something like "Feature extraction is the problem of obtaining..." or something like "Feature extraction consists in obtaining..."
 - line 233, please, check the sentence "For a simple linear process that data". It should probably be something like "For a simple linear process in which data..."
 - please, "et al." is an abbreviation of the Latin expressions "et alii" ("and others"). The full stop must therefore be put. Check the "Kano et al" at line 270, the "Cheng et al" at line 277, "Zhano et al" at line 693, line 796, line 833, line 835, line 895, and maybe there are others.
 - line 299, "it's" is too colloquial. It should be "it is".
 - line 353, "hid-den" should be "hidden"
 - line 416, "to controlled just" should probably be "to be controlled just"

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Thanks to the authors for their response to the review. I am happy with the additions because they put more emphasis on questions that still need to be answered within the research into AI applications in the process industry. My last comment/question is to use a few sentences in the recommendations or the conclusions to pay attention to the question of what such a development means for the knowledge development of the operators about the use and application of these new technological possibilities. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx