Bayesian Survival Analysis Model for Girth Weld Failure Prediction

Round 1

Reviewer 1 Report

The proposed model can be fitted using the maximum likelihood approach as well.

Is there any advantages on the Bayesian approach in this model over the maximum likelihood approach?

The novelty of the paper rests in the three stages model to incorporate the 14 failure factors. If no one has done this model, it is better to state in the paper.

Author Response

Dear referee:

     Thank you very much for your comments and suggestions on our paper applsci-462550. 

We have revised the manuscript according to your kind advices and referee’s detailed suggestions.

Enclosed please find the responses to the referees. We also improved other content,especially for the result and conclusion presented. PLS tell me if you have any question.

Thank you very much for all your help and looking forward to hearing from you soon. 

Comment 1：

The model can also be fitted by  the maximum likelihood method based on the current sample, regardless of the a prior information. But The result also present the situation what is happening at the moment. The Bayesian model is to obtain the posterior distribution to predict the probability under different cases through the prior distribution  and the likelihood function. We aim to predict the change trend of girth weld with different factors. And in the paper we also compare the classic Kalan-Meier survivals model and Bayesian model. The advantages on the Bayesian is the applicability to predict  the probability considering many factors.

 Comment 2：

The Staged Bayesian failure model for the girth weld is the first time proposed in this paper.Thank you for your remind, we have stated this in the revised version.

Best regards

Author Response File: Author Response.docx

Reviewer 2 Report

This paper presents a staged-survival model for predicting girth weld failure. It used a linear model to account for the hazardous factors related to weld failure.

The major issue of the article is inadequate description of the training and validation datasets.

1.      Besides the 2 events (weld accidents) in the validation dataset, it is not clear how many negative controls are in the validation dataset (i.e. cases with risk factors but no events). A more comprehensive description of the validation dataset seems necessary for this paper.

2.      It seems arbitrary that 14 out of 64 accidents are used in training.

3.      It could be important to provide more justification on the use of staged-survival model. The selection of cutoff years seems arbitrary. Is the model supported by previous studies? How is its model fitting statistics compared to other models?

 

Author Response

Dear referee:

I am very grateful to your comments for the manuscript. We have revised the manuscript according to your kind advices and referee’s detailed suggestions. Enclosed please find the responses to the referees. We sincerely hope this manuscript will be finally acceptable to be published on Applied Sciences. Thank you very much for all your help and looking forward to hearing from you soon.

Sincerely yours

comment 1:The procedure actually is a risk assessment method, only analyzing the potential probability of girth weld failure based on the accident happened and information. We can not have the conclusion that the weld with a low survival probability would be failure,but the potential is very high according the historical information.This also the limit for risk assessment. I have found some shortcomings in my current work due to your suggestion. It is necessary to consider the detailed parameters in the model using strength and fracture theory.But it is very complex, and I will keep on the research in future.  Only the two failure cases with the factor external force caused by soil movement that had a great impact on the result.This is also consistent with the situation on site.We have revised the description of the validation data.

comment 2:The Staged Bayesian survival model with 3 stages was proposed in section 2.3 according all the 64 accidents data, mainly considering the life time.After that we also need to know what the  influence of different factors on the failure,especially for "infant mortality" stage, that is, the new pipelines (mentioned in the introduction,the failure of girth weld is the main risk for new pipeline).Otherwise,we invested and collected the detailed failure information for the 14 accidents , which can be used to analyze.And we only have limited data for other accident happened many years ago.But the procedure we proposed in section 3 can be applied to other stages based on sufficient information（failure report or causes）.

comment 3:Based on the frequency of girth weld failure and Bathtub Curve of pipeline,we also divided the survival curve into 3 stages named  "infant mortality" failures, the constant failure and the "wear out" failures.The cutoff year between stages is 10 and 40, which was selected according to the historical statistics and turning point in the survival curve of girth weld,which is described in section 2.2.We also compare the classic Kalan-Meier survivals model and Staged Bayesian model. The advantages on the Bayesian is the applicability to predict  the probability considering many factors.The Staged Bayesian failure model for the girth weld is the first time proposed in this paper.  This is an attempt to set up a framework method of girth weld prediction with Bayesian survival analysis algorithm and improve the traditional risk assessment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Thank you for addressing my comments.