Research On The Reliability Analysis Method And System Of Agricultural Tractors Based On Warranty Data

Tractors are indispensable power machines for agricultural mechanization.Researches on the reliability analysis method of tractors are of great significance for improving the quality of tractors and promoting the development of agricultural mechanization.By far,researches on the reliability analysis of tractors were mainly based on reliability tests or follow-up surveys,and those reliability data collected by tests can hardly have a balance between volume and dimension.Warranty data,which are collected by enterprises when providing warranty services and handling other related businesses,are a valuable and easily accessible data source for manufacturers to assess the reliability of products in the field.Warranty data cover the whole life cycle of products,and reliability analysis based on warranty data can help enterprises achieve closed-loop product quality management.In this work,based on the warranty data of tractors,researches on the reliability analysis method of tractors are carried out from the aspects of estimating the reliability function,establishing the reliability model with covariates,and establishing the competitive failure model with covariates.Post hoc interpretation for machine learning models and association rule mining are applied to explore the important reliability factors and their effects.Finally,based on these works,a reliability analysis system based on warranty data for tractors is designed and developed.This work broadens the data sources and methods available for reliability analysis of tractors,establishes a new reliability analysis method based on warranty data for tractors,and provides special tools for enterprises to utilize warranty data to realize the intelligent analysis of products’reliability.The main research contents and conclusions of this work are as follows:（1）A field reliability estimation method of tractors based on warranty data is proposed and applied to estimate the field reliability of six types of tractors.Based on the many years’warranty data from a tractor manufacturer,the proposed method establishes a usage rate regression model（RMSE=2.35 h/d）to impute the usage of unfailed tractors,and estimates the field reliability based on the pseudo-complete usage dataset.By using the reliability estimation results to predict the number of warranty claims in the next eight months for validation,compared with the other two baseline methods,the proposed method’s prediction accuracy increases by 16.4%on average,indicating that the proposed method can estimate the field reliability of tractors better.（2）The random survival forest（RSF）model is used to establish models with 34 potential reliability factors,which are extracted from the warranty data and related to the production and operation of tractors,and important factors and their effects are explored by post hoc interpretation.The effects of some important factors are also transformed into the decision rules for identifying high-risk tractors to better serve the tractor enterprises.The results show that the RSF models,which consider the observed time as the age and operation hour of tractors respectively,have close performance（C-index is 0.88 and 0.83,IBS is 0.09 and 0.15,respectively）and show similar factor importance and effect.The results also shows that the usage rate is the most important factor for tractors’reliability.According to the effect of other important factors,some potential failure patterns are found:Heavty-duty tractors have higher failure hazards（HR=1.43,p-value<0.001）;tractors used in central regions have lower failure hazards（HR=1.38,p-value<0.001）.（3）Competitive failure models with reliability factors are established by taking the seven most common types of failures for tractors as competitive failure events.Cause-specific Cox proportional-hazards model,Fine-Gray model,RSF,and Deep Hit are applied for modeling.The non-failure-specific factors and failure-specific factors for the seven types of failures are further explored by post hoc interpretation.In the five-fold cross-validation,the overall performance of the RSF model and Deep Hit model for the seven types of failures are better(C^td-index is 0.79 and0.74 respectively).The non-failure-specific factors include the usage rate and rated horsepower of tractors;the failure-specific factors include tractors’s special model type.According to the effect of the failure-specific factors,some failure-specific failure patterns are found.For example,the component provided by the supplier C5S2 can increase the failure hazard of hydro-cylinder（cs-HR=1.82,sub-HR=1.86,p-value<0.001）;The font axle of special type tractors,such as paddy field type,are more likey to fail（cs-HR=3.14,sub-HR=2.96,p-value<0.001）.（4）The Apriori algorithm is used to mine association rules in the warranty data to further explore the failure pattern of tractors.Support,confidence,lift,Kulczynski,and imbalance ratio are used as the interest measures to filter interesting rules at multiple levels,and the reliability of the tractor matched by these rules is estimated.By interpreting the results from different aspects,some potential risks that will increase the failure rate were found.For example,the supplier combination,which is represented by an itemset with 79.70%support,makes the tractors’reliability lower than the average level;tractors used in eastern regions are more likely to have hitch failures;tractors are more likely to have the same type of failures after their first failures.（5）A reliability analysis system based on warranty data for tractors is designed and developed.According to the relevant design standard,the three-layer architecture,which includes a data layer,a service layer,and an application layer,of the system is determined.Based on the above analyses of the warranty data,the function modules of the system are designed,including“field reliability estimation”,“reliability factor analysis”,“competitive failure analysis”and“association rules analysis”.Finally,a prototype of the system is developed and deployed by using the Shiny framework and Docker.The system improves the capability of tractor manufacturers for analysis and utilization of warranty data,and support them for making more effective production management decision plans.