MK-DCCA-Based Fault Diagnosis for Incipient Faults in Nonlinear Dynamic Processes
Round 1
Reviewer 1 Report
The proposed article is interesting, and the theoretical background of the implemented model is complete and adequately described.
There are, however, a few clarifications regarding the application of the model (from paragraph 4 onwards), which appear necessary in order to better represent its usability in real world situations.
1. randomly generated dataset. It is required to make explicit the criteria of the generative algorithm, and to clarify its choices.
2. CSTR Simulink simulation. it is required to define the parameters of the simulator in order to identify, or hypothesise, the deviations that led to the selected faults.
3. in general, since the faults are memory-less processes, it is required to comment on the evolution of the systems studied over time, so as to hypothesise critical modes and variables, complementing the outcomes of the model.
4. Concerning point 3, it might be appropriate to compare the implemented model with possible applications of other approaches, such as HMMs (see, as an example, the paper https://doi.org/10.1016/j.psep.2023.02.058), which could introduce useful insights regarding the early detection of faults.
a grammar chech is suggested.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
This paper is devoted to fault diagnosis and identification of incipient faults in dynamical processes. To solve it, a modified Principal Components Analysis method was proposed, developed and named Mixed Kernel Principal Components Analysis and Dynamic Canonical Correlation Analysis (MK-DCCA). The comparison of the proposed algorithm with the existing ones showed better performance, that proved its workability.
Before considering this paper for possible publication in Processes journal several issues should be clarified and corrected:
1. In (1), it is not clear if the kernel function $\Phi$ is taken of 1 or 2 arguments. What is its sense if it is a function of 1 argument?
2. In (2), is it a scalar product of x and x'?
3. Line 114: explain, why it is necessary to perform centering.
4. In (7), what kind of operation is denoted by ( , )?
5. Line 138: what does it mean "opts"?
6. Line 164: the last component in $P_u$ should be a misprint.
7. Line 172: it is not clear how fault detection in dynamical is resolved.
8. Line 274: is there any algorithm to chose the number of singular values?
9. In (51), are all the undefined notations the same as in (10) and (13)?
10. In 4.1.2, it is said that the model was trained on a fault-free dataset. Is such a training correct? How the model "catches" the faults then?
11. Were the metrics from (52) used for comparison? There is no information in the paper.
12. In numerical results, it is not clear if and how the system reacts on singular spikes, crossing the treshhold. (See, for example, Figure 4 e and f).
Moderate English language editing is required.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
After reading the manuscript, I could observe that approached subject is: the incipient faults diagnosis by means of a new statistical method, namely, a combined method of Mixed Kernel Principal Components Analysis and Dynamic Canonical Correlation Analysis (MK-DCCA). The basic idea of this method involves combining the RBF kernel (Gaussian kernel functions) and polynomial kernel to form a mixture kernel through a weighted mathematical relationship.
Such situations can appear in modern process industry systems leading to degraded system performance, abnormal shutdowns, and even catastrophic consequences.
Two cases are studied: In Case I, the proposed method is first applied to a randomly generated dataset to demonstrate its robust generalization performance (the process is assumed to have measurement white noise). Case II utilizes the method on a simulated model of a continuous stirred tank reactor (CSTR) using a Simulink model (the process is assumed to have measurements without noise and measurement white noise).
The results of the case studies are shown and graphically explained, demonstrating its excellent detection performance for incipient fault in nonlinear and dynamic system of the MK-DCCA method. The performance indicators for monitoring of different faults using different methods are presented by tables. The advantages and disadvantages of the applied method are highlighted.
Some corrections are needed from text editing point of view, as examples:
1) Page 3 / line 111: in Eq.2 Kpoby is Kpoly?
2) Page 4 / line 131: “ . . . j=1, . . . , N. Then . . . “ instead of “ . . . j=1, . . . , N, Then . . . “
3) Page 9 / Eq (43) : Cine este Uf ?
4) Page 11 / on the Figure 1 on the axes labeled on the top: “space” instead of “spcae”
5) Pages 14, 16, 17 / all Figures: “ T2 ” instead of “ T2 ”
6) Page 18 / line 428: “ Ci ” instead of “ Ci ”
7) Page 18 / line 430: who is Tci ?
Please review carefully these errors.
Title is informative and reflects the contents and the Introduction chapter reflects a good documentation. The conclusions pointing out the relevance of some obtained results but could be improved by the intention for future studies.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
The problem of detecting incipient faults in industrial process systems is discussed and a modified CCA process monitoring method is proposed.
The research topic is relevant for early detection of failures in industrial processes. It is proposed to take time into account as a parameter.
It is proposed to take time into account as a parameter. It is assumed that in order to develop the proposed methodology, research should be aimed at finding methods for adaptive selection of parameters and calculation of thresholds.
The conclusions in the conclusion section correspond to the presented evidence and arguments and meet the research objective.
Remarks
1. The list of references should be expanded to 35-40 sources.
2. Reference #24 should be clarified.
3. For the first example, the data used to train the model is not presented.
Minor editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
I thank the authors for the responses to the comments, which fully respond to the raised doubts.
Reviewer 2 Report
All the reviewers questions have been answered, corresponding modifications have been made, and the paper can be considered for possible publication.
Moderate changes are recommended.