Product field failure data association rule mining method

Abstract

The invention discloses a product field failure data association rule mining method, which comprises the following steps: analyzing product after-sales service text information, making an assumption for repair service text information, establishing failure criteria of a product and components thereof, designing a field failure data mining process, and constructing an association rule-based mining method. Evaluating the effectiveness of the association rule-based mining method, mining and statistical analysis of actual association failure data, and finally applying a mining result. The invention provides a field failure data association rule mining method oriented to mass product after-sales service text information formed by enterprise accumulation, so as to efficiently obtain a high-quality field failure data set. The method can be applied to field reliability prediction and warranty period design of products and components thereof, and can also provide an important basis for after-sales operation and maintenance decisions of enterprises. The dependence on manpower is reduced, and the data acquisition efficiency and the data quality are ensured.

Description

Technical Field

[0001] The present invention relates to a field failure data processing technology, which is applicable to after-sales service text information for field reliability evaluation and prediction of products and their components, and in particular to a method for mining association rules of product field failure data. Background Art

[0002] The field failure data of enterprise products and their components truly reflects their field reliability level, and the after-sales service text information contains extremely rich field failure data of products and their components. Therefore, the study proposes field failure data for after-sales service text information and its mining technology based on association rules to efficiently obtain high-quality field failure data sets, which is not only necessary for the reliability design and evaluation prediction of products and their components, but also provides an important basis for enterprise after-sales operation and maintenance decision-making.

[0003] At present, the massive after-sales service text information accumulated by enterprises is mainly used to calculate the workload of after-sales service and statistically calculate the repair rate of products. Only a limited number of studies use human methods to mine field failure data from it and apply it to the prediction of product field reliability and warranty period design. However, relying on human resources to mine field failure data from after-sales service text information not only has low data acquisition efficiency, but also makes it difficult to ensure data quality. Therefore, in order to mine field failure data efficiently and with high quality from massive after-sales service text information, it is necessary to study and propose scientific mining methods. Summary of the invention

[0004] The purpose of the present invention is to provide a method for mining association rules of product field failure data in order to overcome the defects of the above-mentioned prior art.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] The present invention provides a method for mining association rules of product field failure data, comprising the following steps:

[0007] Obtaining product after-sales service text information, the after-sales service text information includes product operation instructions, maintenance consultation and repair service text information, the field failure data of the product and its components are contained in the repair service text information in the after-sales service text information, the repair service text information includes: the geographical location of the product being repaired, the specific dates of production, installation and maintenance, the specific components that failed and their failure phenomenon and a description of the repair process;

[0008] Making assumptions about the repair service text information;

[0009] Establish failure criteria for the product and its components according to the specific product, wherein the failure includes: associated failure and non-associated failure, and the failure criteria include: failure determination criteria, associated failure determination criteria and non-associated failure determination criteria;

[0010] Performing on-site failure data mining on after-sales service text information according to the failure criteria to obtain a failure information set, a correlated failure information set, a non-correlated failure information set, and a multi-truncated on-site failure data set;

[0011] Establish a product failure dictionary based on the word frequency statistical analysis results of the repair description related content in the repair service text information;

[0012] Analyze the specific clauses of the invalidation information set, the associated invalidation information set and the non-associated invalidation information set one by one, extract words, phrases or short sentences with clear and complete connotations, and use the extracted information as the key constituent elements of the corresponding association rules;

[0013] By analyzing the dependency syntax of the repair service text information, combined with the key components of the association rules and the failure dictionary of the product, mining or judgment rules for failure information, associated failure information and non-associated failure information, namely the association rules, are formulated respectively;

[0014] According to the association rules and field failure data mining process, a field failure data mining method program is constructed and a computer mining algorithm is written;

[0015] Based on the multiple truncated field failure data set, a point estimate of the field failure rate of the product or its components is obtained by using a maximum likelihood point estimation method;

[0016] According to the multiple truncated field failure data set, a maximum likelihood interval estimation method is used to obtain an interval estimation value of the field failure rate of the product or its components.

[0017] Furthermore, through further analysis, the product repair description content classifies the reasons for on-site failure of products and their components into three categories: the first category is design or manufacturing defects in the products and their components; the second category is insufficient guidance on product installation or use by after-sales installers; and the third category is customer misoperation, false repair reports and force majeure.

[0018] Furthermore, the failure refers to: the product or its components lose the specified function, the performance parameters exceed the specified range or the structure is damaged.

[0019] Furthermore, the assumptions include: the repair date of the product or its components is its expiration date, the product installation date is its commissioning date, the recorded failures are all failures that occurred during the use of the product, when the product and its components fail, the customer will report the failure to the company's after-sales service department, and the information will be fully recorded, and all warranty repairs are regarded as the first failure report of the product or its components.

[0020] Furthermore, the associated failure refers to failure of a product or its components during use due to design or manufacturing defects.

[0021] Furthermore, the non-associated failure refers to the failure of the product or its components during use due to improper after-sales installation and debugging, improper operation or force majeure.

[0022] Furthermore, the field failure data mining process includes two binary classification information mining and one information statistical analysis.

[0023] Furthermore, the two binary classification information mining and one information statistical analysis include: classifying the after-sales service text information based on the failure judgment criteria of the failure judgment criterion to obtain a failure information set, classifying the failure information set based on the associated failure judgment criteria and the non-associated failure judgment criteria in the failure judgment criterion to obtain an associated failure information set and an unassociated failure information set, and according to whether the field reliability assessment is for products or components, and whether it is divided into geographical regions or not, for each piece of information in the associated failure information set, according to the number of products installed within each unit's installation time, analyzing and counting the number of field failures and truncated numbers of products or specific components within each unit's working time to obtain its multiple truncated field failure data set.

[0024] Furthermore, the effectiveness evaluation method of the field failure data association rule mining method includes:

[0025] Apply the on-site failure data mining method based on association rules to the binary information mining stage of failure information and non-failure information, and obtain the failure information set N2 by mining the after-sales service text information test set N1;

[0026] Applying the field failure data mining method based on association rules to the binary classification information mining stage of associated failure information and non-associated failure information, mining the failure information set N2 to obtain the associated failure information set N3;

[0027] The effectiveness of the association rule mining method is evaluated by using three evaluation indicators: the accuracy P, the recall R and the harmonic mean F1 of the information classification and mining results. The formula is as follows:

[0028]

[0029] Among them, TP is the number of information whose true category is a positive example and whose mining result is also a positive example; FP is the number of information whose true category is a negative example and whose mining result is a positive example; FN is the number of information whose true category is a positive example and whose mining result is a negative example.

[0030] Furthermore, the computer mining algorithm applies the field failure data mining method program based on association rules according to the established association rules and mining process, determines the failure information, the association rules of associated failure information and non-associated failure information, and the after-sales service text information set N1, and then performs syntactic dependency analysis on the information in the after-sales service text information set N1 one by one, and then determines whether it belongs to failure information according to the failure determination association rules, and determines whether the failure information belongs to associated failure information according to the associated failure determination association rules, and finally outputs the failure information set N2 and the associated failure information set N3.

[0031] Compared with the prior art, the present invention has the following advantages:

[0032] The present invention proposes a method for mining field failure data association rules for the massive product after-sales service text information accumulated by enterprises over time, so as to efficiently obtain high-quality field failure data sets. It can be applied to the prediction of field reliability of products and their components and the design of warranty periods, and can also provide an important basis for the after-sales operation and maintenance decision-making of enterprises. It reduces the dependence on manpower and ensures data acquisition efficiency and data quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a program diagram of the field failure data mining method based on association rules of the present invention;

[0034] Figure 2 It is a flow chart of field failure data mining based on failure criteria in the present invention;

[0035] Figure 3 This is an example diagram of the dependency syntax analysis of the maintenance description in the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0037] Embodiment 1:

[0038] In view of the problems existing in the field failure data mining of the above-mentioned product after-sales service text information, this embodiment provides a method for mining association rules of product field failure data, which is used to efficiently obtain high-quality field failure data sets, provide data support for the field reliability design and evaluation prediction of products and their components, and provide a basis for the company's after-sales operation and maintenance decisions.

[0039] The method for mining association rules of product field failure data has the following steps:

[0040] 1) Analyze product after-sales service text information:

[0041] 1.1 The products described include: the products and their components described in the after-sales service text information;

[0042] 1.2 The after-sales service text information includes product operation instructions, maintenance consultation and repair service content, and the generated after-sales service text information set is recorded as N1;

[0043] 1.3 The field failure data of the product and its components is contained in the repair service text information in the after-sales service text information;

[0044] The repair service process described in 1.4 is as follows: the user dials the service hotline - calls to make an appointment - on-site inspection of the fault - quotation - customer confirmation - start of repair - completion of repair - user confirmation;

[0045] 1.5 After completing the repair, the maintenance personnel fill out the repair form and upload it to the company's after-sales service system to form a repair service text information;

[0046] 1.6 Analysis of the enterprise repair service text information shows that its content includes: the geographical location of the repaired product, the specific date of production, installation and implementation of the repair, the specific components that failed and their failure phenomenon and a description of the repair process;

[0047] 1.7 From the composition of the repair service text information, it can be known that: the geographical location of the product being repaired, that is, the specific place where the product is used; the installation date and the specific date of the repair are the basis for determining the use or operation time of the product; the repair description is the description of the repair service content completed by the maintenance personnel, including the relevant description of the confirmation, repair and verification process of the on-site failure of the product and its components;

[0048] 1.8 Through further analysis of the product repair description, the reasons for the on-site failure of the product and its components are classified into three categories: the first category is the design or manufacturing defects of the product and its components; the second category is the inadequate guidance of the after-sales installers on product installation or use; the third category is customer misoperation, false repair reports and force majeure;

[0049] 1.9 Among the above reasons for repairs, only the field failures caused by the first category of reasons reflect the attribute characteristics of the field reliability of the product and its components, and belong to the research scope of field reliability; according to the specific content of the repair description, the text information of the repair service caused by the first category of reasons is comprehensively and accurately mined;

[0050] 2) The following assumptions are made regarding the repair service text information:

[0051] 2.1 The repair date of the product or its components shall be its expiration date;

[0052] 2.2 The date of product installation is the date of its commissioning;

[0053] 2.3 The recorded failures are all those that occurred during the use of the product;

[0054] 2.4 When the product and its components fail, customers will report the failure to the company's after-sales service department, and the information will be fully recorded;

[0055] 2.5 Since the number of repair reports and the repair intervals of the product and its components are not recorded, this repair report is regarded as the first failure repair report of the product or its components;

[0056] 3) Establish failure criteria for the product and its components:

[0057] 3.1 Failure as described above means: the product or its components lose their specified functions, their performance parameters exceed the specified range, or their structure is damaged;

[0058] 3.2 The failures described include: associated failures and non-associated failures;

[0059] 3.3 The associated failure mentioned above refers to the failure of the product or its components during use due to design or manufacturing defects;

[0060] 3.4 The non-associated failure mentioned above refers to the failure of the product or its components during use due to improper after-sales installation and commissioning, improper operation or force majeure;

[0061] 3.5 The failure criteria described above are established based on specific products, including: failure judgment criteria, associated failure judgment criteria and non-associated failure judgment criteria;

[0062] 4) Design field failure data mining process:

[0063] 4.1 The on-site failure data mining process based on the failure criteria is essentially a process of classifying the after-sales service text information;

[0064] 4.2 In order to reduce the difficulty and complexity of information category analysis and judgment, the entire mining process is designed to consist of three sequential stages: two binary classification information mining and one information statistical analysis;

[0065] 4.3 The task of the first binary classification information mining stage is to mine the after-sales service text information set N1 to obtain the failure information set N2: based on the failure judgment criteria in the failure judgment criteria, determine whether the information in N1 is failure information one by one, and divide the after-sales text information into two categories: failure information and non-failure information. The collection of all failure information constitutes the failure information set N2;

[0066] 4.4 The task of the second binary classification information mining stage is to mine the failure information set N2 to obtain the associated failure information set N3: according to the associated failure judgment criteria and the non-associated failure judgment criteria in the failure judgment criteria, the failure information in N2 is mined and analyzed one by one, that is, the category to which it belongs is determined, and the failure information is divided into two categories: associated failure information and non-associated failure information. The collection of all the information belonging to the associated failure category constitutes the associated failure information set N3;

[0067] 4.5 The information statistical analysis stage is performed on the associated failure information set N3: according to whether the field reliability assessment is for products or components, and whether it is divided into geographical regions or not (provinces, cities, counties), for each piece of information in the associated failure information set N3, according to the number of products installed within the installation time of each unit, the field failure number and truncated number of products or specific components within the working time of each unit are analyzed and counted, and the multiple truncated field failure data set N is obtained;

[0068] 5) Construct a mining method based on association rules:

[0069] 5.1 Establish a product failure dictionary based on the word frequency statistical analysis results of the repair description related content in the repair service text information;

[0070] 5.2 Analyze the specific clauses of the failure, associated failure and non-associated failure judgment criteria that constitute the failure judgment criteria one by one, and extract words, phrases or short sentences with clear and complete connotations, which are the key components of the corresponding association rules;

[0071] 5.3 By analyzing the dependency syntax of the repair service text information, combined with the key components of the association rules and the failure dictionary of the product, mining or judgment rules for failure information, associated failure information and non-associated failure information, namely the association rules, are formulated respectively;

[0072] 5.4 According to the association rules and field failure data mining process, construct a field failure data mining method program and write a computer mining algorithm;

[0073] 6) Evaluate the effectiveness of the association rule mining method:

[0074] 6.1 The on-site failure data mining method based on association rules is applied to the binary information mining stage of failure information and non-failure information, and the failure information set N2 is obtained by mining the after-sales service text information test set N1;

[0075] 6.2 Apply the field failure data mining method based on association rules to the binary classification information mining stage of associated failure information and non-associated failure information, and obtain the associated failure information set N3 from the failure information set N2;

[0076] 6.3 The effectiveness of the association rule mining method is evaluated by using three evaluation indicators: the accuracy P, the recall R and the harmonic mean F1 of the information classification and mining results. The formula is as follows:

[0077]

[0078] Where: TP is the number of information whose true class is a positive example and whose mining result is also a positive example; FP is the number of information whose true class is a negative example and whose mining result is a positive example; FN is the number of information whose true class is a positive example and whose mining result is a negative example;

[0079] 7) Mining and statistical analysis of actual correlation failure data:

[0080] 7.1 Apply the mining method based on association rules that has passed the effectiveness evaluation to the mining of the actual after-sales service text information set to obtain the associated failure information set;

[0081] 7.2 According to the needs of field reliability estimation, statistical analysis is performed on the associated failure information set to obtain a multi-truncated field failure data set of the product or its components in the geographical area (province, city, county) where the product is located;

[0082] 8) Application of mining results:

[0083] 8.1 Based on the multiple truncated field failure data set, use the maximum likelihood point estimation method to obtain a point estimate of the field failure rate of the product or its components;

[0084] 8.2 Based on the multiple truncated field failure data set, a maximum likelihood interval estimation method is used to obtain an interval estimate of the field failure rate of the product or its components.

[0085] Preferably, the computer mining algorithm is based on the established association rules and mining process, applies the field failure data mining method program based on association rules, determines the failure information, the associated failure information and the non-associated failure information association rules, and the after-sales service text information set N1, and then performs syntactic dependency analysis on the information in the after-sales service text information set N1 one by one, and then determines whether it belongs to the failure information according to the failure determination association rules, and determines whether the failure information belongs to the associated failure information according to the associated failure determination association rules, and finally outputs the failure information set N2 and the associated failure information set N3.

[0086] Embodiment 2:

[0087] The parts not mentioned in this embodiment are the same as those in Embodiment 1.

[0088] The present invention is further described below in conjunction with an embodiment of a household air conditioner.

[0089] The specific implementation steps of the on-site failure data association rule mining method for after-sales service text information of household air-conditioning products are as follows:

[0090] Step 1: Analyze the after-sales service text information of household air conditioners:

[0091] Table 1 shows some entries of after-sales repair service text information of a certain model of household air conditioner derived from the after-sales service text information set of a certain air conditioner company.

[0092] Table 1 Example of after-sales repair text information for a certain model of household air conditioner

[0093]

[0094]

[0095] From the composition of the repair service text information, we can know that: the customer address, that is, the specific place where the air conditioner is used, is the basis for analyzing the geographical differences in the field reliability of the air conditioner and its components and their influencing factors; the installation date and maintenance date are the basis for determining the specific use or operation time of the air conditioner; the maintenance description is the description of the repair service content completed by the after-sales service personnel in response to the specific problems reported by the user, which generally includes the relevant description of the confirmation, repair and verification process of the field failure of the air conditioner and its components. Through further analysis of the tens of thousands of air conditioner maintenance descriptions currently available, the reasons for the repair can be classified into three categories: one is the design or manufacturing defects of the air conditioner itself; the second is the improper installation of the air conditioner by the after-sales installer or insufficient use guidance; the third is the customer's misoperation, false repair and force majeure. Only the field failure caused by the first type of reasons reflects the attribute characteristics of the field reliability of the air conditioner and its components, and belongs to the research scope of field reliability. Therefore, the field failure data mining for after-sales service text information is to comprehensively and accurately mine the after-sales repair information belonging to the first category according to the specific content of the maintenance description.

[0096] Step 2: Make the following assumptions about the repair service information:

[0097] (1) The date of maintenance of the air conditioner is the expiration date;

[0098] (2) The date the air conditioner is installed is the date it is put into use;

[0099] (3) All recorded failures occurred during the use of the air conditioner;

[0100] (4) When the air conditioner fails, the customer will report the problem to the company's after-sales service department and the repair information will be fully recorded;

[0101] (5) Since there is no record of how many times the air conditioner has been repaired or the interval between repairs, all reported failures are considered to be its first failure;

[0102] Step 3: Establish the failure criteria of the air conditioner:

[0103] As an example, Table 2 gives the failure judgment criteria for air conditioners and some of their components, and Table 3 gives the judgment criteria for air conditioner-associated failure and non-associated failure of components.

[0104] Table 2 Failure judgment criteria for air conditioners and their components

[0105]

[0106] Table 3 Criteria for determining associated failure and non-associated failure of air-conditioning components

[0107]

[0108] Step 4: Design the field failure data mining process:

[0109] The entire mining process is designed to consist of three sequential stages: two binary classification information mining and one information statistical analysis (e.g. Figure 2 As shown), that is, the following classification and judgment are performed on each piece of information in the air conditioner after-sales service text information set N1201: first, the classification and judgment of failure information and non-failure information, that is, according to the failure judgment criteria 205 of the air conditioner and its components, determine whether the information is failure information, if not, then classify and judge the next piece of information; if yes, then include it in the failure information set N2202, and perform the following classification and judgment on the information as to whether it is associated failure information or not; second, the classification and judgment of associated failure information and non-associated failure information, that is, according to the associated failure and non-associated failure judgment criteria 206 of the air conditioner components described in Table 3, determine whether the information determined as failure is associated failure information, if yes, then include it in the associated failure information set N3203; after completing the classification and judgment of all information in the air conditioner after-sales service text information set N1, according to the requirements of the on-site reliability assessment, analyze and count the associated failure information of the air conditioner or specific components to obtain its on-site failure data set N204.

[0110] Step 5: Construct the key elements and association rules of association rules:

[0111] Based on the word frequency statistical analysis results of the repair description related content in the air conditioner repair service text information, the failure dictionary of household air conditioners shown in Table 4 is established;

[0112] Table 4 Failure dictionary of household air conditioners

[0113]

[0114] The specific clauses of the failure judgment criteria of air conditioners and some of their components and the judgment criteria of associated failure and non-associated failure shown in Tables 2 and 3 are analyzed one by one, and then words, phrases or phrases with clear and complete connotations are extracted, and the key components of the corresponding association rules (AR) shown in Table 5 are obtained;

[0115] Table 5 Key components of association rules (AR)

[0116]

[0117]

[0118] Dependency parsing techniques are applied to identify the dependencies between words, phrases, and words in the maintenance description sentences. For example, Figure 3The syntactic analysis results of a sentence "After on-site inspection and replacement of the outdoor unit mainboard and the indoor unit mainboard" in the description of air conditioner maintenance in Table 1 are given by using the Jieba word segmentation system; through syntactic analysis, not only the interdependence between words in the sentence is identified, such as the parallel relationship between the words "on-site" and "inspection", the subject-object relationship between "replacement" and "mainboard", and the attributive relationship between "outdoor unit" and "mainboard", but also the syntactic structure of the sentence is determined, thus laying a foundation for data mining.

[0119] Based on the household air conditioner failure dictionary in Table 4, the key AR components in Table 5, and the dependency syntactic analysis results of after-sales repair service information, the mining or judgment rules of failure information, associated failure information, and non-associated failure information, namely AR, can be formulated respectively, as shown in Table 6. Each AR is responsible for the mining of the information judged by its corresponding failure judgment criterion;

[0120] Table 6 AR set for mining failure information and associated failure information

[0121]

[0122]

[0123] According to the established association rules and mining process, apply Figure 1 The field failure data mining method program based on association rules is shown, and the computer mining algorithm is written in Python 3.7 programming language, and is implemented by running Pycharm software. The computer reads the failure dictionary of the household air conditioner described in Table 4, the association rules for determining failure information, associated failure information and non-associated failure information described in Table 6, and the air conditioner after-sales service text information set N1, and then performs syntactic dependency analysis on the information in the after-sales service text information set N1 one by one, and then determines whether it belongs to failure information according to the failure determination association rules, and determines whether the failure information belongs to associated failure information according to the associated failure determination association rules, and finally outputs the failure information set N2 and the associated failure information set N3.

[0124] Step 6: Evaluate the effectiveness of the association rule mining method:

[0125] Table 7 shows the test set N1 of after-sales service text information of four models of household air conditioners used to evaluate the effectiveness of association rules, where the test set of type I air conditioner is randomly generated from its after-sales service text information set, and the test sets of type II, III and IV air conditioners are their complete after-sales service text information sets. According to the field failure data mining process, the failure information set N2 and the associated failure information set N3 (see Table 7) are obtained by processing with human mining and computer mining algorithms based on association rule mining methods.

[0126] Table 7 Test set of after-sales service text information of four types of air conditioners and their human mining results

[0127]

[0128] Considering that the quality of the mining results can be continuously improved through multiple iterations of the human mining process, and the final mining results that tend to be stable can be regarded as the true value of the field failure data, the human mining results are used as the comparison benchmark of the proposed association rule mining method. Table 8 gives the mining results and evaluation index values ​​(%) of the association rule mining method for four models of household air conditioners.

[0129] Table 8 Mining results and evaluation index values ​​obtained by association rule mining method (%)

[0130]

[0131] From the results in Tables 7 and 8, we can see that since the proportion of failure information in the after-sales service text information test set N1 is as high as 97.0%-97.6%, and the proportion of non-failure information is extremely small, showing a serious skewed distribution, the P, R and F1 values ​​of the failure data N2 mining results of the association rule mining method are generally better than the corresponding values ​​of the failure data N3 mining results. However, the P, R and F1 values ​​of the four types of air conditioners are not much different, and the fluctuation amplitude is also very small.

[0132] From the evaluation index values ​​(%) of the final mining results of the associated failure data N3 of the two currently mainstream text information neural network mining methods and the association rule mining method, namely, the gated recurrent unit (GRU) and the capsule network (CAP), given in Table 9, it can be seen that the three index values ​​of P, R and F1 of the associated failure data N3 obtained by the proposed association rule mining method are much better than the corresponding index values ​​of the mining results obtained by the GRU and CAP mining methods, and the fluctuation amplitudes of the three index values ​​are generally better than the fluctuation amplitudes of the corresponding index values ​​of the mining results obtained by the GRU and CAP mining methods, that is, it has better robustness.

[0133] Table 9 Evaluation index values ​​of N3 of associated failure data mining results obtained by different mining methods (%)

[0134]

[0135] From the results in Table 8 and Table 9, we can see that the established mining method based on association rules can obtain high-quality mining results and is robust, and can be effectively applied to the mining of associated failure information in actual after-sales service information sets.

[0136] Step 7: Mining and statistical analysis of actual correlation failure data:

[0137] According to the on-site failure data mining process, the aforementioned mining method based on association rules that has passed the effectiveness evaluation is applied to the mining of the complete after-sales service text information set of the four types of air conditioners, and the failure and associated failure information sets shown in Table 10 are obtained.

[0138] Table 10 After-sales service text information set of four types of air conditioners and their association rule mining results

[0139]

[0140] Taking the cooling (April to September of each year) or heating (October to March of the following year) cycle time of the air conditioner as one working unit time and one installation unit time, statistical analysis is performed on the mined associated failure information set N3, and a multiple truncated field failure data set of air conditioners or their components with or without the geographical area (province, city, county) where the air conditioner is located can be obtained. As an example, Table 11 gives a multiple truncated field failure data set of outdoor electric control components of four types of air conditioners regardless of the geographical area where the air conditioner is located.

[0141] Table 11 Multiple truncated field failure data set of outdoor electronic control components of four types of air conditioners

[0142]

[0143]

[0144] Step 8: Application of mining results:

[0145] According to the multiple truncated field failure data set shown in Table 11, the maximum likelihood point estimation or interval estimation method can be used to calculate the point estimate and interval estimate of the field failure rate of the outdoor electronic control components of the four types of air conditioners shown in Table 12, with confidence intervals of 90% and 95% respectively.

[0146] Table 12 Estimated values ​​of on-site failure rates of outdoor electric control components of four types of air conditioners

[0147]

[0148] The field failure data association rule mining method for product after-sales service text information is used to efficiently obtain high-quality field failure data sets suitable for field reliability assessment and prediction of products and their components, providing a basis for enterprises' product design and after-sales operation and maintenance decisions oriented to field reliability.

[0149] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for mining association rules of product field failure data, characterized in that: The following steps are involved: Obtaining product after-sales service text information, the after-sales service text information includes product operation instructions, maintenance consultation and repair service text information, the field failure data of the product and its components are contained in the repair service text information in the after-sales service text information, the repair service text information includes: the geographical location of the product being repaired, the specific dates of production, installation and maintenance, the specific components that failed and their failure phenomenon and a description of the repair process; Making assumptions about the repair service text information; Establish failure criteria for the product and its components according to the specific product, wherein the failure includes: associated failure and non-associated failure, and the failure criteria include: failure determination criteria, associated failure determination criteria and non-associated failure determination criteria; Performing on-site failure data mining on after-sales service text information according to the failure criteria to obtain a failure information set, a correlated failure information set, a non-correlated failure information set, and a multi-truncated on-site failure data set; Establish a product failure dictionary based on the word frequency statistical analysis results of the repair description related content in the repair service text information; Analyze the specific clauses of the invalidation information set, the associated invalidation information set and the non-associated invalidation information set one by one, extract words, phrases or short sentences with clear and complete connotations, and use the extracted information as the key constituent elements of the corresponding association rules; By analyzing the dependency syntax of the repair service text information, combined with the key components of the association rules and the failure dictionary of the product, mining or judgment rules for failure information, associated failure information and non-associated failure information, namely the association rules, are formulated respectively; According to the association rules and field failure data mining process, a field failure data mining method program is constructed and a computer mining algorithm is written; Based on the multiple truncated field failure data set, a point estimate of the field failure rate of the product or its components is obtained by using a maximum likelihood point estimation method; According to the multiple truncated field failure data set, a maximum likelihood interval estimation method is used to obtain an interval estimation value of the field failure rate of the product or its components.

2. The method for mining association rules of product field failure data according to claim 1, characterized in that: After further analysis of the product repair descriptions, the causes of repair reports leading to on-site failure of products and their components are classified into three categories: the first category is design or manufacturing defects in the products and their components; the second category is insufficient guidance on product installation or use by after-sales installers; and the third category is customer misoperation, false repair reports and force majeure.

3. The method for mining association rules of product field failure data according to claim 1, characterized in that: Failure refers to: the product or its components lose their specified functions, performance parameters exceed the specified range or the structure is damaged.

4. The method for mining association rules of product field failure data according to claim 1, characterized in that: The assumptions include: the repair date of the product or its components is its expiration date, the product installation date is its commissioning date, all recorded failures are failures that occur during the use of the product, when the product and its components fail, customers will report the failure to the company's after-sales service department and the information will be fully recorded, and all warranty repairs are regarded as the first failure report of the product or its components.

5. The method for mining association rules of product field failure data according to claim 1, characterized in that: The associated failure refers to the failure of a product or its components during use due to design or manufacturing defects.

6. The method for mining association rules of product field failure data according to claim 1, characterized in that: The non-associated failure refers to the failure of the product or its components during use due to improper after-sales installation and debugging, improper operation or force majeure.

7. The method for mining association rules of product field failure data according to claim 1, characterized in that: The field failure data mining process includes two binary classification information mining and one information statistical analysis.

8. The method for mining association rules of product field failure data according to claim 7, characterized in that: The two binary classification information mining and one information statistical analysis include: classifying after-sales service text information based on the failure judgment criteria of the failure criteria to obtain a failure information set, classifying the failure information set based on the associated failure judgment criteria and the non-associated failure judgment criteria in the failure criteria to obtain an associated failure information set and an unassociated failure information set, and according to whether the field reliability assessment is for products or components, and whether it is divided into geographical regions or not, for each piece of information in the associated failure information set, according to the number of products installed within the installation time of each unit, analyzing and counting the number of field failures and truncated numbers of products or specific components within the working time of each unit to obtain its multiple truncated field failure data set.

9. The method for mining association rules of product field failure data according to claim 1, characterized in that: The effectiveness evaluation method of the field failure data association rule mining method includes: Apply the on-site failure data mining method based on association rules to the binary information mining stage of failure information and non-failure information, and obtain the failure information set N2 by mining the after-sales service text information test set N1; Applying the field failure data mining method based on association rules to the binary classification information mining stage of associated failure information and non-associated failure information, mining the failure information set N2 to obtain the associated failure information set N3; The effectiveness of the association rule mining method is evaluated by using three evaluation indicators: the accuracy P, the recall R and the harmonic mean F1 of the information classification and mining results. The formula is as follows: Among them, TP is the number of information whose true category is a positive example and whose mining result is also a positive example; FP is the number of information whose true category is a negative example and whose mining result is a positive example; FN is the number of information whose true category is a positive example and whose mining result is a negative example.

10. The method for mining association rules of product field failure data according to claim 1, characterized in that: The computer mining algorithm applies the field failure data mining method program based on association rules according to the established association rules and mining process, determines the failure information, the association rules of associated failure information and non-associated failure information, and the after-sales service text information set N1, and then performs syntactic dependency analysis on the information in the after-sales service text information set N1 one by one, and then determines whether it belongs to failure information according to the failure determination association rules, and determines whether the failure information belongs to associated failure information according to the associated failure determination association rules, and finally outputs the failure information set N2 and the associated failure information set N3.

Images