Causal Plot: Causal-Based Fault Diagnosis Method Based on Causal Analysis

Round 1

Reviewer 1 Report

 

1.      Why does multivariate statistical process control (MSPC) not always correctly diagnose the cause of failure?

 

2.      In this study, is the proposed causal plot valid for fault diagnosis problems other than the vinyl acetate monomer (VAM) manufacturing process?

 

3.      There is only one equation for equation (6), equation (7) and equation (8).

 

4.      Why should an equation be written as equation (6), equation (7) and equation (8)?

 

5.      There are 66 process variables in the VM model, which are indicated by circled numbers in Fig. 2 and listed in Table 1. Can the order of the 66 process be adjusted to make it easier to read?

 

6.      The measurement duration of one dataset was 20 hours with 7,200 measurements, the sampling interval of the simulator being 10 seconds. Why is the sampling interval 10 seconds?

 

7.      The measurement duration of one dataset was 20 hours with 7,200 measurements, the sampling interval of the simulator being 10 seconds. The normal and faulty data were defined as 7, 200 × 66 matrixes. Figure 3 shows the fault detection results of MAL1 - MAL4 by MSPC. Can Figure 3 show twenty hours of time?

 

8.      Some content and the paper "Uchida, Y., Fujiwara, K., Saito, T., & Osaka, T. (2022). Process Fault Diagnosis Method Based on MSPC and LiNGAM and its Application to Tennessee Eastman Process. IFAC-PapersOnLine, 55(2), 384-389." is similar, please indicate the citation.

Author Response

The authors greatly appreciate your evaluation and helpful suggestions. Our replies to the comments are as follows.

1. Why does multivariate statistical process control (MSPC) not always correctly diagnose the cause of failure?

We appreciate your comment. Westerhuis et al. [8] have reported the possibility that a fault increases the contributions of process variables unrelated to the fault cause as well as the variables directly related to the fault cause. They concluded that the residuals between the PCA model and the original process data may be computationally distributed to various variables other than the variable related to the fault cause. 

We have cited reference [8] and added the above explanation to the Introduction section in the revised manuscript.

2. In this study, is the proposed causal plot valid for fault diagnosis problems other than the vinyl acetate monomer (VAM) manufacturing process?

Thank you for your valuable suggestion. We reported the application result of the proposed method to the Tennessee Eastman Process, which is widely used as the process benchmark of the fault detection and diagnosis methods, in the previous conference paper [19]. 

In this revision, we emphasized this point at the end of Section 4.4. 
    

3. There is only one equation for equation (6), equation (7) and equation (8).
4. Why should an equation be written as equation (6), equation (7) and equation (8)?

We assume that the relationship among x1-x3 in Fig. 1 is linear. According to Fig. 1, x1 is affected by only e1 x3 is affected by x1 and x2 as well as e3, which are written as Eqs. (6) and (8). In addition, Eqs. (6) - (8) can be rewritten as one equation Eq. (9) when all equations are assumed as linear.
    
    

5. There are 66 process variables in the VM model, which are indicated by circled numbers in Fig. 2 and listed in Table 1. Can the order of the 66 process be adjusted to make it easier to read?

The assigned numbers of the process variables in Fig. 2 and Table 1 were defined in the original VM model [18]. We have added this point to the caption of Fig. 2 of the revised manuscript.

6. The measurement duration of one dataset was 20 hours with 7,200 measurements, the sampling interval of the simulator being 10 seconds. Why is the sampling interval 10 seconds?

  
The sampling interval 10 seconds was defined as the default value of the VM model. We explained this point in Sec. 4.1 of the revised manuscript.

7. The measurement duration of one dataset was 20 hours with 7,200 measurements, the sampling interval of the simulator being 10 seconds. The normal and faulty data were defined as 7, 200 × 66 matrixes. Figure 3 shows the fault detection results of MAL1 - MAL4 by MSPC. Can Figure 3 show twenty hours of time?

 

Figure 3 showed the fault detection results from 20 minutes before to one hour after the fault occurrences of MAL1 - MAL4 by MSPC to emphasize the behaviors around the fault occurrences.

In this revision, we added the figure of the fault detection results for 20 hours with 7,200 measurements as Supplementary Figure S1.

 

8. Some content and the paper “Uchida, Y., Fujiwara, K., Saito, T., & Osaka, T. (2022). Process Fault Diagnosis Method Based on MSPC and LiNGAM and its Application to Tennessee Eastman Process. IFAC-PapersOnLine, 55(2), 384-389.” is similar, please indicate the citation.

We have cited our conference paper as [19] and described as follows:

“A preliminary version of this work has been reported in [19]. In this study, we add a case study of the VAM process and a detailed analysis of the relationship between the causal plot and a recycle flow in processes.

Reviewer 2 Report

(1) Steps to acquire the casual plots should be clearly descried and verified with either experimental or simulated datasets.

(2) How to deal with the casual plots when there is a big size of process variables should be discussed. 

(3) How to determine the number of "S"(uppercase) should be discussed.

Author Response

The authors greatly appreciate your evaluation and helpful suggestions. Our replies to the comments are as follows.

1. Steps to acquire the casual plots should be clearly descried and verified with either experimental or simulated datasets.

We appreciate your valuable indication. We summarized the procedure of the causal plots at the end of Sec. 3 of the revised manuscript.

2. How to deal with the casual plots when there is a big size of process variables should be discussed.

The application of LiNGAM to large datasets was studied in reference [22], and the extension of the causal plots to large process data would be future work. We added future work in Sec. 5 of the revised manuscript.

3. How to determine the number of “S”(uppercase) should be discussed.

In this study, since the causal plots were applied to the process simulation, S was determined 1 hour, following Kanse et al. [31], who reported that it might take about one hour to identify causes of faults in the actual processes. 

We have added the above explanation to Sec. 4.3 of the revised manuscript.

 

Round 2

Reviewer 1 Report

Accept in present form.

Reviewer 2 Report

The authors are suggested to correct many typos and writing mistakes.