JP2014194369A - Component failure occurrence prediction method, component failure occurrence prediction program and component failure occurrence prediction device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve accuracy for prediction of a failure in a component for each vehicle.SOLUTION: In a component failure occurrence prediction device 1, a parameter selection section 21 selects parameters that affect a failure of a predetermined component and regarding the selected parameters, an evaluation model generation section 23 performs ranking learning on the basis of a travel distance of a failure-free vehicle and generates an evaluation model representing a degree of influence on the failure of the component. An evaluation value calculation section 24 then applies the generated evaluation model to calculate the degree of influence regarding a vehicle subjected to calculation of failure occurrence timing of the component and a failure probability calculation section 26 predicts occurrence of the failure of the component on the basis of the calculated degree of influence.

Description

  The present invention relates to a component failure occurrence prediction method and the like.

  Replacement of parts at regular distances or times is performed on the basis of the travel distance of the vehicle and timing standards. However, since the driving conditions differ depending on each vehicle, there is a demand for determining the replacement time for each vehicle in consideration of the vehicle, parts, and driving conditions, instead of changing them uniformly.

  For example, a technique is known in which an estimation model is created based on the causal relationship between various parameters representing running conditions and the occurrence of a component failure, and an appropriate timing is estimated.

  Even if the causal relationship between various parameters representing driving conditions and the occurrence of parts failure is very complex, the driving data of a faulty car that fails within a predetermined driving distance is a positive example, and the driving data of a faultless car is negative. For example, a machine learning technique is known. In such a technique, it is possible to predict a failure of the vehicle by using the traveling data of the vehicle for which the replacement time is to be calculated as an input of the discrimination model.

JP 2007-328522 A International Publication No. 2003/034159

  However, when the replacement time is predicted for each vehicle in consideration of the traveling conditions, there is a problem that it is difficult to improve the prediction accuracy of the failure in the parts for each vehicle. That is, there are very few parts that have a clear causal relationship between various parameters representing driving conditions and the occurrence of a failure. In addition, in conventional machine learning, the travel data for a failed vehicle (positive example) is less than the travel data for a non-failed vehicle (negative example), so appropriate discrimination used to predict vehicle failure It is difficult to create a model. Therefore, it is difficult to improve the failure prediction accuracy of the parts for each vehicle.

  In one side, it aims at improving the prediction accuracy of the failure in the parts for every vehicle.

  In the component failure occurrence prediction method disclosed in the present application, a computer selects a parameter that affects the failure of a predetermined component, and learns a ranking of a travel distance based on the travel distance of a faultless vehicle for the selected parameter. Performing ranking learning, generating an evaluation model representing the degree of influence on the failure of the part, applying the created evaluation model, calculating the degree of influence for a vehicle for which a part failure occurrence time is determined, Based on the calculated degree of influence, each process for predicting the occurrence of a failure of the component is executed.

  It is possible to improve the prediction accuracy of a failure in a part for each vehicle.

FIG. 1 is a functional block diagram illustrating the configuration of the component failure occurrence prediction apparatus according to the embodiment. FIG. 2 is a diagram illustrating an example of a data structure of travel data information. FIG. 3 is a diagram illustrating an example of a data structure of component replacement information. FIG. 4 is a diagram illustrating an example of the data structure of the parameter selection information. FIG. 5 is a diagram illustrating an example of a data structure of parameter application result information. FIG. 6 is a diagram illustrating an example of the distribution of travel data. FIG. 7 is a diagram illustrating an example of a data structure of the evaluation result information. FIG. 8 is a diagram illustrating an example of a survival curve generated by the failure probability calculation unit. FIG. 9 is a diagram illustrating a calculation example of the recommended replacement travel distance. FIG. 10 is a diagram illustrating an example of an output result. FIG. 11 is a flowchart illustrating the evaluation model generation process according to the embodiment. FIG. 12 is a diagram illustrating a flowchart of failure occurrence time prediction processing according to the embodiment. FIG. 13 is a diagram illustrating an example of a computer that executes a component failure occurrence prediction program.

  Hereinafter, embodiments of a component failure occurrence prediction method, a component failure occurrence prediction program, and a component failure occurrence prediction apparatus disclosed in the present application will be described in detail with reference to the drawings. In the following embodiments, a case where the present invention is applied to a component failure occurrence prediction apparatus that predicts the occurrence of a failure of a component mounted on a vehicle will be described. However, the present invention is not limited to the embodiments.

[Configuration of component failure occurrence prediction device]
FIG. 1 is a functional block diagram illustrating the configuration of the component failure occurrence prediction apparatus according to the embodiment. The component failure occurrence prediction device 1 performs ranking learning for learning the order of travel distances given to travel data using only travel data of a fault-free vehicle, and evaluates the degree of influence on the failure according to the travel conditions. An evaluation model is generated. Such processing is referred to as “evaluation model generation processing”. Then, the failure occurrence prediction device 1 obtains an evaluation value based on the generated evaluation model for a vehicle for which a failure occurrence time of a part is to be obtained, and predicts the failure occurrence time using only data close to the obtained evaluation value. . Such processing is referred to as “failure occurrence time prediction processing”. Hereinafter, the failure occurrence prediction apparatus 1 will be specifically described. In the following description, a broken vehicle refers to a vehicle that fails within a predetermined travel distance. A trouble-free vehicle refers to a vehicle that does not break down even when a predetermined travel distance is reached.

  The component failure occurrence prediction device 1 includes a control unit 2 and a storage unit 3.

  The control unit 2 is, for example, an integrated circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA) or an electronic circuit such as a central processing unit (CPU) or a micro processing unit (MPU). Further, the control unit 2 includes a parameter selection unit 21, a parameter selection rule application unit 22, an evaluation model generation unit 23, an evaluation value calculation unit 24, an evaluation result search unit 25, a failure probability calculation unit 26, and an output. Part 27.

  The storage unit 3 corresponds to, for example, a storage device such as a nonvolatile semiconductor memory element such as a flash memory or a FRAM (registered trademark) (Ferroelectric Random Access Memory). Furthermore, the memory | storage part 3 has driving | running | working data DB (DataBase), vehicle management DB, parameter DB, and evaluation result DB. The travel data DB includes travel data information 31. The travel data information 31 stores travel data for each vehicle for each part. The vehicle management DB includes part replacement information 32. In the part replacement information 32, part replacement information is stored for each vehicle and for each part. The parameter DB includes parameter selection information 33 and parameter application result information 34. In the parameter selection information 33, a rule for selecting parameters used for generating the evaluation model is stored for each part. The parameter application result information 34 stores the result of applying the selected parameter for each vehicle and for each part. The evaluation result DB includes evaluation result information 35. In the evaluation result information 35, a result (evaluation value) evaluated using the generated evaluation model is stored for each vehicle and for each part. An example of the data structure of each information will be described later.

  Here, the data structure of the travel data information 31 will be described with reference to FIG. FIG. 2 is a diagram illustrating an example of a data structure of travel data information. As shown in FIG. 2, the travel data information 31 includes a data ID (IDentification) 31a, a vehicle ID 31b, a total travel distance 31c, a section 31d, a stop count 31e, a deceleration distribution 31f to 31i, an air brake operating time 31j, and a retarder. The operating time 31k is included. Data ID31a shows ID which identifies driving data. Vehicle ID31b shows ID which identifies the vehicle corresponding to driving | running | working data. The total travel distance 31c indicates the total travel distance when travel data is stored. The section 31d shows a distinction between failure and no failure. The number of stops 31e and thereafter is information generated from the total result of the sensor information extracted from the vehicle, and corresponds to parameters described later. Here, the information regarding the brake hydraulic device which is one of the components mounted in the vehicle is represented.

  As an example, when the data ID 31a is “1”, the vehicle ID 31b is “1”, the total travel distance 31c is “40.3” (10,000 km), the category 31d is “no failure”, and the stop count 31e is “5. 7 ”(times / 1 km). Further, “0.51”, “0.32”, “0.14”, and “0.03” are stored as the deceleration distributions 31f to 31i, respectively. Further, “30.4” is stored as the air brake operating time 31j, and “10.6” is stored as the retarder operating time 31k.

  The data structure of the parts replacement information 32 will be described with reference to FIG. FIG. 3 is a diagram illustrating an example of a data structure of component replacement information. As shown in FIG. 3, the part replacement information 32 includes a data ID 32a, a vehicle ID 32b, a replacement part name 32c, and a total travel distance 32d during replacement. The data ID 32 a indicates an ID for identifying traveling data, and corresponds to the data ID 31 a of the traveling data information 31. The vehicle ID 32 b indicates an ID for identifying a vehicle corresponding to the travel data, and corresponds to the vehicle ID 31 b of the travel data information 31. The replacement part name 32c indicates the name of the replaced part. The total travel distance 32d during replacement indicates the total travel distance during replacement. The replacement total travel distance 32d indicates the total travel distance from the previous part replacement to the current replacement when the previous part has been replaced.

  As an example, when the data ID 32a is “1”, “1” is stored as the vehicle ID 32b, “brake hydraulic device” is stored as the replacement part name 32c, and “20.8” (10,000 km) is stored as the total travel distance 32d during replacement. ing.

  Returning to FIG. 1, the parameter selection unit 21 uses the travel data of the non-failed vehicle and the travel data of the failed vehicle in the component of interest, and has a property that has a property that has a significant correlation with the total travel distance at the time of failure. Sort out.

  For example, the parameter selection unit 21 searches the travel data information 31 for faulty vehicle data that is travel data in which the classification is faulty. Then, the parameter selection unit 21 searches for the replacement information of the parts in the failed car from the parts replacement information 32 using the searched broken car data. Then, the parameter selection unit 21 selects parameters having a significant correlation with the travel distance (total travel distance during replacement) from the previous part replacement. As an example, when focusing on the brake hydraulic system, the parameter selection unit 21 can select “the number of stops per 1 km” as a parameter that has a significant correlation with the travel distance of the faulty vehicle data.

Moreover, the parameter selection part 21 totals the observed value observed at each time for every predetermined class, when a parameter is represented as distribution. Then, the parameter selection unit 21 calculates a value obtained by linearly combining the observed values collected for each class so as to have the highest correlation with the travel distance of the faulty vehicle data, and the calculated value is significantly different from the total travel distance at the time of the fault. When there is a correlation, the linearly coupled parameters may be selected. As an example, when focusing on the brake hydraulic system, the parameter selection unit 21 determines that the deceleration parameter Σa _i d _i obtained by linearly combining the values d _i of the classes with the coefficient a _i from the frequency distribution of the deceleration is the total travel distance at the time of failure. And whether there is a significant correlation. When it is determined that there is a significant correlation, the parameter selection unit 21 can select the linearly combined parameters and generate an expression represented by the selected parameters and coefficients as a parameter selection rule.

The parameter selection unit 21 can also select a plurality of parameters that are related to each other. As an example, when paying attention to the brake hydraulic system, the parameter selection unit 21 linearizes “air brake operation time around 1 km” B ₁ and “retarder operation time around 1 km” B ₂ with coefficients b ₁ and b ₂ , respectively. It is determined whether or not the combined parameter b ₁ B ₁ + b ₂ B ₂ has a significant correlation with the total travel distance at the time of failure. When it is determined that there is a significant correlation, the parameter selection unit 21 can generate a parameter selection rule including the linear combination coefficients b ₁ and b ₂ .

  As described above, for example, there is a “non-correlation test” as a method for determining whether or not there is a significant correlation with the total travel distance at the time of failure. In addition, although it demonstrated that the parameter selection part 21 selects the parameter which has a property which has a significant correlation with the total mileage at the time of a failure, it is not limited to this, distribution is significant with a failure-free vehicle and a failure vehicle. Parameters with different properties may be selected. As described above, for example, there is a “two-sample Kolmogorov-Smirnov test” as a method for determining whether the distribution is different between the non-failed vehicle and the failed vehicle.

  Then, the parameter selection unit 21 records the selected parameters and the parameters and coefficients expressed by the linear combination formula in the parameter selection information 33 as parameter selection rules. Here, the data structure of the parameter selection information 33 will be described with reference to FIG. FIG. 4 is a diagram illustrating an example of the data structure of the parameter selection information. As shown in FIG. 4, the parameter selection information 33 includes a rule ID 33a, target parameters 33b-1 to 33, and coefficients 33c-1 to 33c. The rule ID 33a indicates an ID for identifying the parameter selection rule. That is, for one rule ID 33a, a parameter that is likely to affect the failure rate of the component is stored together with the coefficient. The target parameter 33b is a parameter that constitutes a parameter selection rule. The coefficient 33c indicates a coefficient corresponding to the target parameter 33b constituting the parameter selection rule. That is, when the parameter selection rule is linear combination, the parameters constituting the linear combination correspond to the parameter 33b, and the coefficients constituting the linear combination formula correspond to the coefficient 33c.

  As an example, when the rule ID 33a is “1”, “the number of stops around 1 km” is stored as the target parameter 33b-1, and “1” is stored as the coefficient 33c-1. That is, the parameter selection rule when the rule ID 33a is “1” is represented by the parameter name itself. When the rule ID 33a is “2”, coefficients “0.1” and “0.2” for each “deceleration distribution” as target parameters 33b-1 to 4 and coefficients “c” for each “deceleration distribution” as coefficients 33c-1 to 33c-4. ”,“ 0.5 ”, and“ 0.2 ”. That is, the parameter selection rule when the rule ID 33a is “2” is represented by a linear combination formula including several parameters of “deceleration distribution”.

  Returning to FIG. 1, in the evaluation model generation process, the parameter selection rule application unit 22 applies the parameter selection rule and generates an application result as an input when generating the evaluation model. For example, the parameter selection rule application unit 22 uses the travel data information 31 and the part replacement information 32 according to the parameter selection rule indicated by each rule ID of the parameter selection information 33, and the value corresponding to each rule ID for each data ID. Is calculated. Then, the parameter selection rule application unit 22 records the value calculated according to each parameter selection rule for each data ID in the parameter application result information 34. Note that the parameter selection rule application unit 22 also generates a parameter application result using travel data and component replacement data of the vehicle that evaluates the component of interest by applying the parameter selection rule even in the failure occurrence time prediction process. To do.

  As an example, a case where a value corresponding to the rule ID 33a “1” of the data ID 31a “1” is calculated according to the parameter identification rule indicated by the rule ID 33a “1” will be described. Since the target parameter 33b-1 is “the number of stops around 1 km”, the parameter selection rule application unit 22 reads the value “5.7” corresponding to the “number of stops around 1 km” 31e from the travel data information 31. Then, the parameter selection rule applying unit 22 calculates a value “5.7” obtained by multiplying the read value “5.7” by the coefficient 33c-1 “1” as a value corresponding to the rule ID 33a “1”. To do. Then, the parameter selection rule application unit 22 records the value corresponding to the rule ID “1” in the parameter application result information 34.

  As another example, a case will be described in which a value corresponding to the rule ID 33a “2” of the data ID 31a “1” is calculated according to the parameter identification rule indicated by the rule ID 33a “2”. Since the target parameter 33b-1 to 4 is "deceleration distribution", the parameter selection rule applying unit 22 determines that the values "0.51" and "0.0" corresponding to the "deceleration distribution" 31f to i from the travel data information 31. “32”, “0.14”, and “0.03” are read out. The parameter selection rule application unit 22 then sums the values obtained by multiplying the read values by the values of the coefficients 33c-1 to 4 (0.51 × 0.1 + 0.32 × 0.2 + 0.14 × 0. 5 + 0.03 × 0.2 =) “0.191” is calculated as a value corresponding to the rule ID 33a “2”. Then, the parameter selection rule application unit 22 records the value corresponding to the rule ID “2” in the parameter application result information 34.

  Here, the data structure of the parameter application result information 34 will be described with reference to FIG. FIG. 5 is a diagram illustrating an example of a data structure of parameter application result information. As shown in FIG. 5, the parameter application result information 34 includes a data ID 34a, a travel distance 34b, a section 34c, a rule ID [1] 34d, a rule ID [2] 34e, and a rule ID [3] 34f. The data ID 34 a indicates an ID for identifying traveling data, and corresponds to the data ID 31 a of the traveling data information 31.

  The travel distance 34b indicates the travel distance of the vehicle from the previous replacement of the part of interest to the present. As an example, it is assumed that the focused component is a “brake hydraulic device”. When the data ID is “1”, the total travel distance 31c of the travel data information 31 is “40.3” (10,000 km), and the total travel distance during replacement corresponding to the “brake hydraulic device” of the parts replacement information 32 32d is “20.8” (10,000 km). Therefore, the travel distance 34b is a value “19.5” obtained by subtracting “20.8” from “40.3”. That is, the travel distance “19.5” (10,000 km) from the previous replacement of the part to the present is stored. For example, the travel distance 34b is calculated and recorded by the parameter selection rule application unit 22.

  The section 34 c indicates a distinction between failure and no failure, and corresponds to the section 31 d of the travel data information 31. The rule ID [1] 34d is a value corresponding to the rule ID 33a “1” in the parameter selection information 33, that is, a value calculated according to the parameter identification rule indicated by the rule ID 33a “1”. The rule ID [2] 34e is a value corresponding to the rule ID 33a “2” in the parameter selection information 33, that is, a value calculated according to the parameter identification rule indicated by the rule ID 33a “2”. The rule ID [3] 34f is a value corresponding to the rule ID 33a “3” in the parameter selection information 33, that is, a value calculated according to the parameter identification rule indicated by the rule ID 33a “3”.

  As an example, when the data ID 34a is “1”, “19.5” is stored as the travel distance 34b, and “no failure” is stored as the section 34c. Further, “5.7” is stored as the rule ID [1] 34d, “0.191” is stored as the rule ID [2] 34e, and “20.5” is stored as the rule ID [3] 34f.

  Returning to FIG. 1, the evaluation model generation unit 23 uses the no-failure vehicle data of the parameter application result information 34 to perform ranking learning in the order of the mileage given to the no-failure vehicle data, so that the component failure An evaluation model that evaluates the degree of impact on is generated. Non-failed vehicle data means running data in which the classification is no failure. Since the ranking learning method itself is an existing technology, a detailed method of ranking learning will not be described.

  Here, the reason why the ranking learning is focused on the travel distance of the failure-free vehicle data will be described. The travel distance of the fault-free vehicle data does not directly indicate the degree of influence on the failure of the component of interest. However, it can be said that the fact that there is no failure even though the vehicle travels a long distance is highly likely that the traveling data includes a traveling condition that is difficult to malfunction. This is apparent from the fact that a vehicle traveling under conditions that are likely to break down cannot travel long distances without failure and is likely to break down. On the other hand, trouble-free vehicle data traveling over a short distance includes various driving conditions from driving conditions that are likely to fail to driving conditions that are less likely to fail. Such characteristics of the travel data are shown in FIG. FIG. 6 is a diagram illustrating an example of the distribution of travel data.

  As shown in FIG. 6, in the travel data distribution graph, the horizontal axis represents the degree of load, and the vertical axis represents the distribution density. That is, this graph shows how much the traveling data of the short traveling distance and the traveling data of the long traveling distance include the traveling conditions for taking the respective load degrees. The degree of load here refers to the degree of load applied to a component during traveling due to the difference in traveling conditions, and is synonymous with the degree of influence on the failure of the component. The travel data for a short travel distance includes a travel condition with a low load level, that is, a failure condition, and also includes data on a travel condition with a high load level, that is, a breakdown condition. On the other hand, the travel data of a long travel distance includes more data of travel conditions with a low load than travel conditions with a high load.

  The ranking learning performed using the travel data having such characteristics is machine-learned based on the premise that travel data with a longer travel distance is more likely to be travel data with a lower load. Although travel data with a low load is also included in travel data with a short travel distance, this is not a problem. As shown in FIG. 6, it can be expected that the ratio of the travel data under the low load condition travel data is smaller than the travel data with a short travel distance compared to the travel data with a long travel distance. It is possible to appropriately learn driving conditions with low load as a feature often seen in data.

  In the evaluation model generation process, the evaluation value calculation unit 24 applies the evaluation model to the failure-free vehicle data and the failure vehicle data in the parameter application result information 34, and calculates an evaluation value respectively. Then, for each data ID, the evaluation value calculation unit 24 records the evaluation value output by the evaluation model, the failure / no-fault distinction, and the travel distance as evaluation results in the evaluation result information 35. Note that regarding failure-free vehicle data, data used for generating an evaluation model and data for calculating an evaluation value may be distinguished.

  Here, the data structure of the evaluation result information 35 will be described with reference to FIG. FIG. 7 is a diagram illustrating an example of a data structure of the evaluation result information. As shown in FIG. 7, the evaluation result information 35 includes a data ID 35a, a travel distance 35b, a section 35c, and an evaluation value 35d. The data ID 35 a indicates an ID for identifying the traveling data, and corresponds to the data ID 31 a of the traveling data information 31 and also corresponds to the data ID 34 a of the parameter application result information 34. The travel distance 35 b indicates the travel distance of the vehicle from the previous replacement of the part of interest to the present, and corresponds to the travel distance 34 b of the parameter application result information 34. The section 35 c shows a distinction between failure and no failure, and corresponds to the section 34 c of the parameter application result information 34. The evaluation value 35d indicates the result of applying the evaluation model. That is, the evaluation value 35d is a value output from the evaluation model as a result of inputting a record for the data ID 34a of the parameter application result information 34 to the evaluation model. For example, when the travel distance is ranked in descending order, the evaluation value indicates a larger value as the load degree is higher. As an example, when the data ID 35a is “1”, “19.5” is stored as the travel distance 35b, “no failure” is stored as the section 35c, and “−1.4” is stored as the evaluation value 35d.

  Returning to FIG. 1, in the failure occurrence time prediction process, the evaluation value calculation unit 24 applies an evaluation model to a parameter application result of a vehicle that evaluates a target component, and calculates an evaluation value.

  The evaluation result search unit 25 searches the evaluation result information 35 for a predetermined number of records having an evaluation value close to the evaluation value calculated for the vehicle that evaluates the component of interest. The evaluation result search unit 25 has been described as searching for a predetermined number. However, the present invention is not limited to this, and a record included in a predetermined evaluation value section including the calculated evaluation value may be searched. As an example, if the evaluation value is 2.0 and the evaluation value section is an evaluation value ± 10%, the evaluation result search unit 25 searches for records having an evaluation value of 1.8 to 2.2.

  The failure probability calculation unit 26 calculates a failure rate corresponding to the travel distance using the travel distance set in each record searched by the evaluation result search unit 25 and the failure / no failure classification. The failure probability calculation 26 generates a survival curve based on the calculated failure rate corresponding to the travel distance. The survival curve can be automatically generated by applying an existing survival analysis method.

  Here, the survival curves obtained for each of the case where the evaluation value is bad (the load degree is high) and the case where the evaluation value is good (the load value is low) are shown with reference to FIG. FIG. 8 is a diagram illustrating an example of a survival curve generated by the failure probability calculation unit. As shown in FIG. 8, it can be seen that the survival curve with a high load degree has a shorter travel distance in the case of the same survival rate than the survival curve with a low load degree. In other words, it can be seen that a vehicle with a high load level needs to replace a focused component at an early stage in order to secure a failure probability comparable to that of a vehicle with a low load level.

  Returning to FIG. 1, the failure probability calculation unit 26 uses the generated survival curve to calculate a recommended replacement distance for the component of interest. For example, the failure probability calculation unit 26 calculates a travel distance in which the survival probability is a predetermined value or less as a recommended replacement travel distance. In other words, the travel distance at which the failure rate is equal to or higher than a predetermined value is calculated as the recommended replacement travel distance. Each of the predetermined values indicates a probability that replacement of a part is recommended, for example.

  Here, an example of calculating the recommended replacement distance for the component of interest using the survival curve will be described with reference to FIG. FIG. 9 is a diagram illustrating a calculation example of the recommended replacement travel distance. In this case, it is assumed that a travel distance in which the survival rate is 90% or less in the survival curve is a recommended replacement travel distance. As illustrated in FIG. 9, the failure probability calculation unit 26 can calculate approximately 400,000 km as a recommended replacement travel distance for a vehicle having a poor evaluation value (high load) using a survival curve. The failure probability calculation unit 26 can calculate about 550,000 km as a recommended replacement travel distance for a vehicle with a good evaluation value (low load) using a survival curve.

  From the survival curve generated by the failure probability calculation unit 26, the output unit 27 outputs, for example, a travel distance at which the failure rate is a predetermined value or more to a monitor. In other words, the failure probability calculation unit 26 outputs, for example, a travel distance at which the survival rate is a predetermined value or less to a monitor. The output unit 27 predicts the travel distance from the registration date and the current travel distance of the vehicle to be evaluated for the component of interest to the next inspection, and the probability of failure by the next inspection based on the predicted result. And the calculated failure probability may be output to a monitor, for example. In other words, the target vehicle assumed to be a service according to the embodiment is a business vehicle such as a transport truck, and it can be assumed that how these vehicles are used in daily work will not change greatly in the future. Easy and accurate prediction is possible. Further, the output unit 27 determines whether or not the failure rate corresponding to the travel distance expected by the next periodic inspection is equal to or higher than a predetermined value. In other words, the output unit 27 determines whether or not the travel distance predicted until the next periodic inspection reaches a travel distance that becomes the recommended replacement travel distance. And the output part 27 may monitor-output the determination result, for example.

  Here, an example of the output result will be described with reference to FIG. FIG. 10 is a diagram illustrating an example of an output result. As shown in FIG. 10, the part name d1, the failure probability d2 until the next inspection, and the replacement recommendation flag d3 are shown as output results. The component name d1 is the component name of the component of interest. The failure probability d2 until the next inspection is the probability of failure until the next inspection. The replacement recommendation flag d3 is a flag indicating whether or not replacement is recommended. Here, “required” is displayed when replacement is recommended, and “unnecessary” is displayed when replacement is not recommended. For example, the output unit 27 outputs “necessary” when it is determined that the travel distance predicted by the next periodic inspection reaches the travel distance that is the recommended replacement travel distance. On the other hand, the output unit 27 outputs “unnecessary” when it is determined that the travel distance predicted by the next periodic inspection does not reach the travel distance that is the recommended replacement travel distance.

  As an example, when the part name d1 of the part of interest is “brake hydraulic device”, “6.2”% is displayed on the monitor as the failure probability d2 until the next inspection, and “necessary” is displayed as the replacement recommendation flag d3.

[Procedure for evaluation model generation processing]
Next, the procedure of the evaluation model generation process will be described with reference to FIG. FIG. 11 is a flowchart illustrating the evaluation model generation process according to the embodiment. It is assumed that the component name of the component of interest is input.

  First, the parameter selection unit 21 acquires travel data from the travel data information 31 (step S11). For example, the parameter selection unit 21 acquires failure vehicle data and non-failure vehicle data from the travel data information 31.

  And the parameter selection part 21 acquires the replacement information of the components to which attention is paid in the failed vehicle from the component replacement information 32 using the failed vehicle data (step S12).

  Then, the parameter selection unit 21 selects parameters having a significant correlation with the travel distance (total travel distance at the time of failure) since the previous part replacement. The parameter selection unit 21 generates a parameter selection rule represented by the selected parameters and coefficients. The parameter selection unit 21 records the generated parameter selection rule in the parameter selection information 33 (step S13).

  Subsequently, the parameter selection rule application unit 22 applies the parameter selection rule (step S14), and generates an application result as an input when generating the evaluation model. For example, the parameter selection rule application unit 22 uses the travel data information 31 and the part replacement information 32 according to the parameter selection rule indicated by each rule ID of the parameter selection information 33, and the value corresponding to each rule ID for each data ID. Is calculated. Further, the parameter selection rule application unit 22 uses the travel data information 31 and the part replacement information 32 to calculate the travel distance of the vehicle from the previous replacement of the component of interest. Then, the parameter selection rule application unit 22 records the generated application result in the parameter application result information 34.

  Subsequently, the evaluation model generation unit 23 extracts trouble-free vehicle data from the parameter application result information 34 (step S15). And the evaluation model production | generation part 23 performs the ranking learning which makes the mileage provided to the extracted failure-free vehicle data in order, and produces | generates the evaluation model which evaluates the influence degree to the failure of components (step S16). .

  Subsequently, the evaluation value calculation unit 24 applies the generated evaluation model to the failure-free vehicle data and the failure vehicle data of the parameter application result information 34, and calculates an evaluation value (step S17). Then, the evaluation value calculation unit 24 records the calculated evaluation value, the failure / no failure distinction, and the travel distance as evaluation results in the evaluation result information 35 (step S18).

  In this way, the evaluation model generation process performs ranking learning using the travel distance in order using only failure-free vehicle data for the component of interest, and generates an evaluation model that evaluates the degree of influence on the failure of the component. .

[Procedure for failure occurrence time prediction]
Next, the procedure of failure occurrence time prediction processing will be described with reference to FIG. FIG. 12 is a diagram illustrating a flowchart of failure occurrence time prediction processing according to the embodiment. It is assumed that traveling data of a vehicle that is to be evaluated for the component of interest is input.

  First, the parameter selection rule application unit 22 obtains travel data of a vehicle that evaluates a target component (step S21). Then, the parameter selection rule application unit 22 acquires vehicle part replacement data for evaluating the part of interest (step S22). Then, the parameter selection rule application unit 22 applies the parameter selection rule using the acquired travel data and part replacement data (step S23), and generates an application result.

  Subsequently, the evaluation value calculation unit 24 applies an evaluation model to the application result and calculates an evaluation value (step S24). Then, the evaluation result search unit 25 searches the evaluation result information 35 for a predetermined number of evaluation results close to the calculated evaluation value (step S25). Then, the failure probability calculation unit 26 calculates a failure probability corresponding to the travel distance by using the travel distance set for each evaluation result searched for a predetermined number and the failure / no failure classification. Then, the failure probability calculation unit 26 generates a survival curve based on the calculated failure probability corresponding to the travel distance (step S26).

  Then, the failure probability calculation unit 26 uses the generated survival curve to calculate a recommended replacement distance for the relevant part of the vehicle that evaluates the part of interest. Then, the output unit 27 outputs the calculated recommended replacement travel distance to, for example, a monitor (step S27).

  In this way, the failure occurrence time prediction process can generate an appropriate evaluation model for each component of interest even when there is little failure vehicle data, so calculate the recommended replacement distance for the vehicle you want to evaluate for the component of interest. it can.

[Effect of Example]
According to the above embodiment, the component failure occurrence prediction device 1 selects parameters that affect the failure of a predetermined component. Then, the component failure occurrence prediction device 1 performs ranking learning on the selected parameters based on the travel distance of the non-failed vehicle, and generates an evaluation model that represents the degree of influence on the component failure. Then, the component failure occurrence prediction device 1 applies the generated evaluation model, calculates the degree of influence on the vehicle for which the failure occurrence time of the component is obtained, and generates the occurrence of the component failure based on the calculated degree of influence. Predict. According to such a configuration, the component failure occurrence prediction device 1 generates the evaluation model that represents the degree of influence on the failure of the component using only the data of the non-failed vehicle. Failure prediction accuracy can be improved. As a result, the component failure occurrence prediction device 1 can calculate an appropriate replacement timing in consideration of different driving conditions for each vehicle regarding the component failure occurrence timing, that is, the replacement timing of the component. Furthermore, the component failure occurrence prediction apparatus 1 can realize appropriate preventive maintenance from which unnecessary part replacement costs are eliminated.

  In addition, according to the above-described embodiment, the component failure occurrence prediction device 1 generates an evaluation model in which a failure-free vehicle having a long travel distance has a higher ratio of travel conditions that are less likely to fail than a vehicle having a short travel distance. To do. According to such a configuration, the component failure occurrence prediction device 1 can improve the failure prediction accuracy of the component in consideration of the traveling conditions for each vehicle.

[Programs]
In the above-described embodiment, the evaluation model generation unit 23 uses the no-failure vehicle data to perform ranking learning in the order of the mileage given to the no-failure vehicle data, and determines the degree of influence on the failure of the component. It has been explained that an evaluation model to be evaluated is generated. However, the evaluation model generation unit 23 is not limited to this, and sets the appropriate threshold value to distinguish between the short-distance traveling data and the long-distance traveling data depending on whether or not the threshold is exceeded, so that the two ranks are learned. Anyway. In addition, the evaluation model generation unit 23 sets two appropriate thresholds, the data of the travel distance not exceeding the smaller threshold 1 is the short distance travel data, and the data of the travel distance exceeding the larger threshold 2 is the long You may make it perform learning of two orders | ranks as distance driving | running | working data. Whichever method is used, an evaluation model that evaluates the degree of influence on the failure of parts (loading degree of driving conditions) by machine learning focusing on the driving distance using only a large amount of trouble-free vehicle data. Can be generated.

  Moreover, in the said Example, the component failure generation | occurrence | production prediction apparatus 1 demonstrated as what estimates generation | occurrence | production of the failure of the components mounted in the vehicle. However, the target for which the occurrence of a component failure is predicted by the component failure occurrence prediction device 1 may be not only a vehicle but also a ship or an airplane. In other words, the target may be any object for which traveling data is generated by the sensor.

  Further, the component failure occurrence prediction apparatus 1 can be realized by mounting each function such as the control unit 2 and the storage unit 3 on an information processing apparatus such as a known personal computer or workstation.

  In addition, each component of the illustrated apparatus does not necessarily need to be physically configured as illustrated. In other words, the specific mode of device distribution / integration is not limited to that shown in the figure, and all or part of the device is functionally or physically distributed / integrated in arbitrary units depending on various loads and usage conditions. Can be configured. For example, the failure probability calculation unit 26 and the output unit 27 may be integrated as one unit. On the other hand, the failure probability calculation unit 26 may be distributed into a function unit that calculates a failure probability and a function unit that generates a survival curve. The parameter application rule application unit 22 and the evaluation value calculation unit 24 may be distributed into a function unit for evaluation model generation processing and a function unit for failure occurrence time prediction processing, respectively. Further, the storage unit 3 may be connected as an external device of the component failure occurrence prediction device 1 via a network.

  The various processes described in the above embodiments can be realized by executing a program prepared in advance on a computer such as a personal computer or a workstation. Therefore, in the following, an example of a computer that executes a component failure occurrence prediction program that realizes the same function as the component failure occurrence prediction apparatus 1 illustrated in FIG. 1 will be described. FIG. 13 is a diagram illustrating an example of a computer that executes a component failure occurrence prediction program.

  As illustrated in FIG. 13, the computer 200 includes a CPU 203 that executes various arithmetic processes, an input device 215 that receives input of data from the user, and a display control unit 207 that controls the display device 209. The computer 200 also includes a drive device 213 that reads a program and the like from a storage medium, and a communication control unit 217 that exchanges data with other computers via a network. The computer 200 also includes a memory 201 that temporarily stores various types of information and an HDD 205. The memory 201, CPU 203, HDD 205, display control unit 207, drive device 213, input device 215, and communication control unit 217 are connected by a bus 219.

  The drive device 213 is a device for the removable disk 211, for example. The HDD 205 stores a component failure occurrence prediction program 205a and component failure occurrence prediction related information 205b.

  The CPU 203 reads the component failure occurrence prediction program 205a, expands it in the memory 201, and executes it as a process. Such a process corresponds to each functional unit of the component failure occurrence prediction apparatus 1. The component failure occurrence prediction related information 205b corresponds to the travel data information 31, the component replacement information 32, the parameter selection information 33, the parameter application result information 34, and the evaluation result information 35. For example, the removable disk 211 stores information such as the component failure occurrence prediction program 205a.

  Note that the component failure occurrence prediction program 205a is not necessarily stored in the HDD 205 from the beginning. For example, the program is stored in a “portable physical medium” such as a flexible disk (FD), a CD-ROM, a DVD disk, a magneto-optical disk, or an IC card inserted into the computer 200. Then, the computer 200 may read and execute the component failure occurrence prediction program 205a from these.

  The following supplementary notes are further disclosed with respect to the embodiments including the above examples.

(Supplementary note 1)
Select parameters that affect the failure of a given part,
For the selected parameters, ranking learning is performed to learn the mileage ranking based on the mileage of a fault-free vehicle, and an evaluation model that represents the degree of influence on the failure of the component is generated.
Applying the generated evaluation model, the degree of influence is calculated for a vehicle for which the failure occurrence time of a part is to be obtained,
A component failure occurrence prediction method, comprising: executing each process for predicting occurrence of a failure of the component based on the calculated degree of influence.

(Additional remark 2) The process which produces | generates the said evaluation model produces | generates the evaluation model in which the ratio which is a driving | running condition which is hard to break down compared with the vehicle with a short driving distance becomes high in the failure-free vehicle with a long driving distance. The part failure occurrence prediction method according to Supplementary Note 1.

(Supplementary note 3) The component failure occurrence prediction method according to supplementary note 1, wherein the parameter selection process includes selecting a parameter having a significantly different distribution between a non-failed vehicle and a failed vehicle.

(Supplementary note 4) The part failure occurrence prediction method according to supplementary note 1, wherein the parameter selection process selects a parameter having a significant correlation with a travel distance at the time of failure.

(Appendix 5)
Select parameters that affect the failure of a given part,
For the selected parameters, ranking learning is performed to learn the mileage ranking based on the mileage of a fault-free vehicle, and an evaluation model that represents the degree of influence on the failure of the component is generated.
Applying the created evaluation model, the degree of influence is determined for a vehicle for which a failure occurrence time of a part is to be obtained,
A component failure occurrence prediction program that executes each process for predicting occurrence of a failure of the component based on the obtained degree of influence.

(Appendix 6) A selection unit for selecting a parameter that affects a failure of a predetermined part;
For the parameters selected by the selection unit, a learning unit that performs ranking learning for learning a mileage ranking based on a mileage of a non-failed vehicle, and a generation unit that generates an evaluation model representing the degree of influence on the failure of the component;
Applying the evaluation model created by the generation unit, a calculation unit that calculates the degree of influence for a vehicle that is a target for determining the failure occurrence time of parts,
A component failure occurrence prediction apparatus comprising: a prediction unit that predicts occurrence of a failure of the component based on the degree of influence calculated by the calculation unit.

DESCRIPTION OF SYMBOLS 1 Component failure occurrence prediction device 2 Control part 3 Storage part 21 Parameter selection part 22 Parameter selection rule application part 23 Evaluation model generation part 24 Evaluation value calculation part 25 Evaluation result search part 26 Failure probability calculation part 27 Output part 31 Traveling data information 32 Parts replacement information 33 Parameter selection information 34 Parameter application result information 35 Evaluation result information

Claims (4)

Computer
Select parameters that affect the failure of a given part,
For the selected parameters, ranking learning is performed to learn the mileage ranking based on the mileage of a fault-free vehicle, and an evaluation model that represents the degree of influence on the failure of the component is generated.
Applying the generated evaluation model, the degree of influence is calculated for a vehicle for which the failure occurrence time of a part is to be obtained,
A component failure occurrence prediction method, comprising: executing each process for predicting occurrence of a failure of the component based on the calculated degree of influence. The process for generating the evaluation model generates an evaluation model in which a failure-free vehicle having a long travel distance has a higher ratio of travel conditions that are less likely to fail than a vehicle having a short travel distance. The part failure occurrence prediction method according to 1. On the computer,
Select parameters that affect the failure of a given part,
For the selected parameters, ranking learning is performed to learn the mileage ranking based on the mileage of a fault-free vehicle, and an evaluation model that represents the degree of influence on the failure of the component is generated.
Applying the created evaluation model, the degree of influence is determined for a vehicle for which a failure occurrence time of a part is to be obtained,
A component failure occurrence prediction program that executes each process for predicting occurrence of a failure of the component based on the obtained degree of influence. A selection unit for selecting parameters affecting the failure of a predetermined part;
For the parameters selected by the selection unit, a learning unit that performs ranking learning for learning a mileage ranking based on a mileage of a non-failed vehicle, and a generation unit that generates an evaluation model representing the degree of influence on the failure of the component;
Applying the evaluation model created by the generation unit, a calculation unit that calculates the degree of influence for a vehicle that is a target for determining the failure occurrence time of parts,
A component failure occurrence prediction apparatus comprising: a prediction unit that predicts occurrence of a failure of the component based on the degree of influence calculated by the calculation unit.

Images