WO2022024453A1 - Failure cause inference device and method - Google Patents

Abstract

In this failure cause inference device that is for inferring a failure cause of a target device and this failure cause inference method that is executed by said failure cause inference device, the target device is configured to: have a plurality of sensors each detecting the state of the target device; store and retain at least one model that has been trained in advance; use a sensor data item in a specific data range that brings a change in the state of the target device from among sensor data items outputted from the sensors; and infer a failure cause of the target device on the basis of the model.

Description

Defect cause estimation device and method

The present invention relates to a defect cause estimation device and a method, and is suitable for application to, for example, a drying defect cause estimation device for estimating the cause of drying defects in a washer / dryer.

Conventionally, when a washer / dryer has a drying defect, it is necessary to identify which of the several possible causes of the defect is the cause. The causes of poor drying include clogging of the filter attached to the intake port for taking in outside air, clogging of the duct attached to the exhaust port, malfunction of the blower, malfunction of the heater, and the like. However, identifying the cause of such a defect is complicated and takes time.

As a technique for determining a defect in a washing machine, Patent Document 1 describes the current applied to the motor and the constant of the motor when the control unit that drives the motor by sensorless control starts the motor by forced commutation. It is disclosed that the rotation speed or the induced voltage of the motor is calculated as a determination parameter, and the rotation failure determination is performed depending on whether or not the determination parameter is in a vibration state.

Further, in Patent Document 2, for example, as a technique for determining the cause of failure of the voltage conversion unit of a motor drive device, after the operation of the voltage conversion unit is stopped, the discharge of the capacitor connected to the output side of the voltage conversion unit is completed. Then, a technique for determining a faulty part based on each voltage value on the input side and the output side of the voltage conversion unit is disclosed.

Japanese Unexamined Patent Publication No. 2019-017179 Japanese Unexamined Patent Publication No. 2005-24567

However, the techniques disclosed in Patent Documents 1 and 2 are determination techniques specialized for failures of specific parts such as motors and voltage converters, and are causes of defects such as filters, ducts, blowers and heaters. There was a problem that it could not be applied to estimate the cause of poor drying.

The present invention has been made in consideration of the above points, and is an attempt to propose a defect cause estimation device and a method capable of efficiently estimating a defect cause.

In order to solve this problem, in the present invention, in the defect cause estimation device for estimating the defect cause of the target device, the target device has a plurality of sensors for detecting the state of the target device, and has been learned in advance. Alternatively, using the storage unit that stores and holds a plurality of models and the sensor data in a specific data range that causes a state change of the target device among the sensor data output from the sensor, the target device is based on the model. A cause estimation unit for estimating the cause of the defect is provided.

Further, in the present invention, in the defect cause estimation method executed by the defect cause estimation device for estimating the defect cause of the target device, the target device has a plurality of sensors for detecting the state of the target device in advance. Using the first step of storing and retaining one or more learned models and the sensor data in a specific data range that causes a state change of the target device among the sensor data output from the sensor, the model is used. Based on this, a second step of estimating the cause of the defect of the target device is provided.

According to the defect cause estimation device and method, it is possible to identify which part of the target device has the defect cause based on the state change of the target device.

According to the present invention, it is possible to realize a defect cause estimation device and a method capable of efficiently estimating the defect cause.

It is a block diagram which shows the schematic structure of the drying defect cause estimation system by 1st Embodiment. It is a block diagram which shows the logical structure of the washing machine and the analysis server by 1st Embodiment. It is a block diagram which shows the structure of the learning system of the cause model. It is a flowchart which shows the learning flow of the cause model. It is a flowchart which shows the flow of cause estimation using a cause model. It is a chart which provides the explanation of the experimental result using CNN as a cause model. It is a chart which provides the explanation of the experimental result using CNN as a cause model. It is a chart which provides the explanation of the experimental result using CNN as a cause model. It is a conceptual diagram provided for the explanation of the 2nd Embodiment. It is a figure which shows the clustering used in the 2nd Embodiment. It is a block diagram which shows the logical structure of the washing machine and the analysis server by 2nd Embodiment. It is a block diagram which shows the structure of the discriminant rule creation system. It is a flowchart which shows the flow of making a discrimination rule. It is a flowchart which shows the flow of the judgment process using the judgment rule. It is a block diagram provided for the explanation of another embodiment.

Hereinafter, one embodiment of the present invention will be described in detail with respect to the drawings.

(1) First Embodiment (1-1) Configuration of Drying Defect Cause Estimating System According to the Present Embodiment In FIG. 1, 1 shows a drying defect cause estimation system according to the present embodiment as a whole. The drying defect cause estimation system 1 is a system having a function of estimating the cause when a drying defect occurs in the washing / drying machine 2, and the washing / drying machine 2 installed in each household and these washing / drying machines are used. The analysis server 3 installed by the manufacturer of the machine 2 is connected to the analysis server 3 via a network 4 such as the Internet.

The washer / dryer 2 is a home electric appliance that washes and dries an object such as clothes according to a user operation, and includes a controller 10, a clock 11, a drive / power unit 12, and various sensors 13.

The controller 10 is a processor that controls the operation of the entire washer / dryer 2. Further, the clock 11 is composed of a digital clock or the like, and has a function as a time counter for ticking the current time and a function as a timer for counting the elapsed time since the start of a certain operation.

The drive / power unit 12 is a washer / dryer 2 such as a motor that rotationally drives a drum in the washer / dryer 2, a blower that blows wind on an object during a drying operation, and a heater that heats the wind output from the blower. It is composed of drive parts and power parts arranged inside.

The sensor 13 is a measuring element that converts the state of the washing / drying machine 2 into a numerical value and outputs it as operation data. For example, a water level sensor for detecting the water level in the washing tub and a water temperature in the washing tub are detected. It is composed of a water temperature sensor for measuring the temperature, a temperature sensor for measuring the temperature of hot air output from the blower, a vibration sensor for measuring the vibration generated by the washing / drying machine 2, and the like.

On the other hand, the analysis server 3 collects operation data output from each sensor of each washer / dryer 2 connected via the network 4 from each washer / dryer 2, and based on the collected operation data, the analysis server 3 collects the operation data. It is a server device having a function of estimating the cause when a drying defect occurs in the washer-dryer 2. The analysis server 3 is composed of a general-purpose server device including a CPU (Central Processing Unit) 20, a memory 21, a storage device 22, a communication device 23, and a display device 24.

The CPU 20 is a processor that controls the operation of the entire analysis server 3. Further, the memory 21 is composed of, for example, a volatile semiconductor memory and is used as a work memory of the CPU 20. The memory 21 stores the cause estimation program 25 and the estimation result output program 26, which will be described later, read from the storage device 22 when the analysis server 3 is started or when necessary.

The storage device 22 is composed of, for example, a large-capacity non-volatile storage device such as a hard disk device or SSD (Solid State Drive), and stores various programs and various data requiring long-term storage. A cause model 27 for each cause of failure, which will be described later as a cause of poor drying, is also stored and held in the storage device 22.

The communication device 23 is composed of a NIC (Network Interface Card) or the like, and performs protocol control during communication with each washer / dryer 2 via the network 4. Further, the display device 24 is composed of, for example, a liquid crystal display or an organic EL (ElectroLuminescence) display, and is used for displaying necessary information.

FIG. 2 shows the logical configuration of the drying defect cause estimation system 1. As shown in FIG. 2, the washer / dryer 2 includes a clock 11, a drive / power unit 12, and various sensors 13, as well as a control unit 30 and a data acquisition unit 31.

The control unit 30 is a functional unit embodied by the controller 10 (FIG. 1) of the washer / dryer 2. The control unit 30 controls the operation of the entire washer / dryer 2 by giving a required command value COM1 to the corresponding drive / power unit 12 at a necessary timing according to a user operation or a program.

Specifically, for example, during the washing operation, the control unit 30 controls the washing operation of the washing / drying machine 2 by appropriately giving a command value specifying the rotation speed and the rotation direction of the drum to the motor. Further, during the drying operation, the control unit 30 controls the drying operation of the washer / dryer 2 by appropriately giving command values such as the wind speed and temperature of the warm air blown to the object to the blower and the heater.

The data acquisition unit 31 is also a functional unit embodied by the controller 10 of the washer / dryer 2. The data acquisition unit 31 is a functional unit having a function of acquiring the command value COM1 output by the control unit 30 to the drive / power unit 12 and the operation data DT1 output from each sensor 13.

At this time, the data acquisition unit 31 selectively acquires the operation data DT1 in a specific data range that causes the state change of the washing / drying machine 2 from the operation data DT1 output from each sensor 13. Specifically, the data acquisition unit 31 selectively selects only the operation data DT1 for a certain period before and after the control unit 30 outputs the command value COM1 to the drive / power unit 12 and / or in a predetermined specific time zone. To get to. Then, the data acquisition unit 31 combines these acquired command values COM1 and operation data DT1 and transmits them to the analysis server 3 via the network 4 (FIG. 1).

On the other hand, the analysis server 3 includes a cause estimation unit 33 including a plurality of determination units 32 and an estimation result output unit 34. The cause estimation unit 33 and the determination unit 32 include a cause estimation program 25 (FIG. 1) in which the CPU 20 (FIG. 1) of the analysis server 3 is stored in the memory 21 (FIG. 1), and a determination program 25A (FIG. 1) which is a part thereof. It is a functional part embodied by executing 1).

The determination unit 32 is provided corresponding to each of the causes of drying defects such as clogging of the filter, clogging of the duct, and malfunction of the blower and the heater. Each determination unit 32 is based on the corresponding operation data DT1 of the various operation data DT1 transmitted from the washer / dryer 2, and the cause model 27 corresponding to the corresponding defect cause given in advance. It is determined whether the drying defect of the washer / dryer 2 is due to the corresponding defect cause, and the determination result is output to the estimation result output unit 34.

The estimation result output unit 34 is a functional unit embodied by executing the estimation result output program 26 (FIG. 1) stored in the memory 21 by the CPU 20 of the analysis server 3. The estimation result output unit 34 sets all the causes of defects determined to be the cause of the drying defect by the corresponding determination unit 32 based on the determination result for each defect cause given by each determination unit 32 as the cause of the drying defect. Is displayed on the display device 24 (FIG. 1) as the estimation result of.

(1-2) Flow of learning of the cause model and flow of abnormality determination using the cause model FIG. 3 shows the flow of learning of the cause model 27 when the cause model 27 is a CNN (Convolutional Neural Network). The "cause model" here is based on the command value COM1 and the operation data DT1 acquired from the washer / dryer 2, and the probability that the cause of the poor drying of the washer / dryer 2 is the corresponding cause is calculated. Refers to a mathematical model that outputs the calculation results.

The washing / drying machine 40 shown in FIG. 3 is a washing / drying machine used for learning the cause model 27, and includes a clock 41, a drive / power unit 42, a plurality of sensors 43, a control unit 44, and a data acquisition unit 45. .. The clock 41, the drive / power unit 42, each sensor 43, the control unit 44, and the data acquisition unit 45 are the corresponding parts (clock 11, drive / power unit 12, sensor 13, control) of the washer / dryer 2 described above with respect to FIG. Since it has the same function and configuration as the unit 30 or the data acquisition unit 31), the description thereof is omitted here.

Similar to the data acquisition unit 31 of FIG. 2, the data acquisition unit 45 of the washing / drying machine 40 has a command value COM2 output by the control unit 44 to the drive / power unit 42 and a command value COM2 of the control unit 44. The operation data DT2 is acquired only for a certain period before and after the output or in a preset time zone, and the acquired command value COM2 and the operation data DT2 are combined to form a learning device via the network 4 (FIG. 1). Output to 50.

The learning device 50 is a computer device provided with information processing resources such as a CPU, a memory, and a storage device (not shown). The learning device 50 may be used in combination with the analysis server 3 (FIG. 1). The learning device 50 is provided with a determination unit 51 and a cause model 27 corresponding to each cause of drying failure such as clogging of a filter, clogging of a duct, and failure of a blower or a heater.

Then, each determination unit 51 is given a command for learning necessary for learning the drying defect caused by the corresponding defect cause among the command value COM2 and the operation data DT2 given by the data acquisition unit 45 of the washing / drying machine 40. The value COM2 and the operation data DT2 are given respectively. For example, the determination unit 51 corresponding to the malfunction of the heater is given a command value COM2 for the heater, operation data DT2 from the temperature sensor that measures the temperature of the air warmed by the heater, and the like. Then, each determination unit 51 inputs the given command value COM2 and the operation data DT2 to the corresponding cause model 27.

In each cause model 27, as shown in FIG. 4, the input command value COM2 and the operation data DT2 are captured via the input layer (S1), and the cause model 27 is based on the captured command value COM2 and the operation data DT2. In the hidden layer (folding layer and pooling layer), the probability that the cause of the drying defect is due to the corresponding defect cause is calculated (S2), and the calculated probability is output from the output layer as a determination result (S3). Then, the determination unit 51 determines whether or not the cause of the drying defect is due to the corresponding defect cause based on the probability output from the cause model 27, and outputs the determination result (S4).

At this time, each determination unit 51 is given an answer as to whether or not the cause of the drying defect is due to the corresponding defect cause based on the user operation or the like for the learning device 50. Thus, the determination unit 51 updates the weight of the filter of the hidden layer of the cause model 27 as necessary based on this answer (S5).

Then, in the learning device 50, each time the command value COM2 and the operation data DT2 are given from the washing / drying machine 40 for learning, the above-mentioned processes of steps S1 to S5 are repeated. As a result, each cause model 27 corresponding to each defect cause is learned.

On the other hand, FIG. 5 shows the flow of the determination process executed by the cause estimation unit 33 of the analysis server 3 using each cause model 27 learned as described above. Each determination unit 32 (FIG. 2) of the analysis server 3 has command values COM1 and operation data DT1 given to each determination unit 32 (FIG. 2) from the data acquisition unit 31 (FIG. 2) of the washer / dryer 2 (FIG. 2) via the network 4. Of these, the command value COM1 and the operation data DT1 according to the corresponding cause are given, respectively. Then, each determination unit 32 inputs the given command value COM1 and the operation data DT1 to the input layer of the corresponding cause model 27 (S10).

In each cause model 27, the input command value COM1 and the operation data DT1 are taken in via the input layer (S11), and based on the taken-in command value COM1 and the operation data DT1, the cause of the drying defect is due to the corresponding defect cause. The probability of being a thing is calculated in the hidden layer (S12), and the calculation result is output from the output layer (S13). Then, the determination unit 32 determines whether or not the cause of the drying defect is due to the corresponding defect cause based on the probability output from the cause model 27, and determines the determination result as the estimation result output unit 34 (FIG. 2). Is output to (S14).

(1-3) Experimental results of cause estimation using CNN Here, FIGS. 6 to 8 show experimental results of cause estimation of drying defects of the washer / dryer 2 performed using CNN as the cause model 27. .. In this experiment, the command values for the blower, heater and flap were used as the command values COM1 and COM2, and the water level sensor and the sensor data of the two temperature sensors were used as the operation data DT1 and DT2. As the cause model 27, a CNN with 3 layers and 38,000 parameters was used.

FIG. 6 shows the experimental results of an experiment in which a part of the duct is clogged in order to cause the duct to dry poorly. In FIG. 6, each numerical value in the state column 70B indicates the ratio of the cross-sectional area of the portion where the duct is clogged to the cross-sectional area of the duct. For example, "0" indicates a state in which the duct is not clogged, and "0.5" indicates a state in which half of the cross-sectional area of the duct is clogged.

Further, the numerical value in the learning number column 70C indicates the number of data sets (number of training data sets) of the command value COM2 and the operation data DT2 in the corresponding state used for learning the cause model 27, and is shown in the verification number column 70D. The numerical value indicates the number of data sets (the number of verification data sets) of the command value COM1 and the operation data DT1 in the corresponding state used for the cause determination process. Further, the numerical value in the correct answer number column 70E indicates the number of verification data sets in which the cause is correct among the number of such verification data sets, and the percentage in the accuracy column 70F indicates the correct answer rate (accuracy) by such verification.

In addition, FIG. 7 shows the experimental results of an experiment in which a part or all of the filter of the intake port is covered with the filter as a cause of poor drying. In FIG. 7, each numerical value in the state column 71B indicates the ratio of the area of the portion covering the filter surface to the entire filter surface of the filter. For example, "0" indicates a state in which the filter surface is not covered at all, and "0.5" indicates a state in which half of the filter surface is covered. Further, "1" indicates a state in which the entire surface of the filter surface is covered. The numerical values and percentage values in the learning quantity column 71C, the verification quantity column 71D, the correct answer number column 71E, and the accuracy column 71F are the same as in FIG.

Furthermore, FIG. 8 shows the experimental results of an experiment in which the exhaust amount of the blower is suppressed so that the failure of the blower is the cause of the poor drying. In FIG. 8, each numerical value in the state column 72B indicates the ratio of the displacement in the experiment to the displacement in the normal state of the blower. For example, "0" indicates a state in which the exhaust amount of the blower is not suppressed, and "0.5" indicates a state in which the exhaust amount of the blower is suppressed to half of the normal state. The numerical values and percentage values in the learning quantity column 72C, the verification quantity column 72D, the correct answer number column 72E, and the accuracy column 72F are the same as in FIG.

In the above experiments, the accuracy of cause estimation due to ducts, filters and blowers is 83% when ducts are the cause, 96% when filters are the cause, and blowers when the cause is blowers. It was 83%. In addition, the time required for estimating the cause was 1 ms or less per washer / dryer regardless of whether the cause was a duct, a filter, or a blower. From this experiment, it was confirmed that practically sufficient estimation accuracy and required time can be obtained by using CNN as the cause model 27.

(1-4) Effect of the present embodiment As described above, in the drying defect cause estimation system 1 of the present embodiment, the control unit 30 of the operation data DT1 output from each sensor 13 of the washer / dryer 2 is used. Defects in the washer / dryer 2 based on the operation data DT1 for a certain period before and after the command value COM1 is output to the drive / power unit 12 and a preset specific time zone, and the cause model 27 for each cause. specify a reason.

Therefore, in the present drying defect cause estimation system 1, it is specified which drive / power unit 12 the control unit 30 has acquired when the command value COM1 is output and the operation data DT1 shows an abnormal value. Along with this, it is possible to identify the cause of poor drying. Therefore, according to the present drying defect cause estimation system 1, it is possible to efficiently estimate the defect cause in which a plurality of possible causes exist.

(2) Second Embodiment (2-1) Method for determining the cause of failure in consideration of operability In a washer / dryer, when the internal air flow during the drying operation is obstructed, the air heated by the heater stays. Occurs, and the temperature distribution in the flow path deviates from the normal state. Therefore, it is considered that the comparison of the measured values of a plurality of temperature sensors is a clue when estimating the cause of the defect.

However, the drying operation is an unsteady process in which drying gradually progresses, and requires complicated control that combines sequence control and feedback control that are completed through multiple steps while constantly estimating the drying state, and the required time is also long. It fluctuates greatly depending on the quantity and quality of the object, outside air temperature, water temperature, etc. In addition, the operating time tends to be extended due to clogging of the filter, which causes poor drying.

For this reason, even if an attempt is made to determine the cause of poor drying in a specific momentary state, not only is it difficult to specify the time to be compared, but there are many other causes that affect the temperature measurement value other than the failure state. It becomes difficult to determine the cause of the defect. For example, in order to estimate a state such as blockage of a flow path from operation data during operation, it is considered necessary to use the characteristics of the dynamic behavior of temperature.

Therefore, in the present embodiment, the time-series data waveform of the operation data is classified (clustered) according to the cause of the drying defect based on its characteristics, and the clustering result is machine-learned to determine the cause of the drying defect (a model for determining the cause of the drying defect (clustering). Hereinafter, this is referred to as a discrimination rule), and the cause of drying failure is estimated using the created discrimination rule.

Here, many studies have been conducted on failure discrimination using time series data, especially on clustering using a similarity scale (dissimilaritymeasure) that quantifies differences and similarities between time series. Much research has been done.

As a typical example, DTW (Dynamic Time Wrapping) has obtained good results in various applications, and various similarity measures have been proposed. Well-known time-series similarity scales can be broadly divided into those based on Euclidean distances and correlations of values and changes (difference series), those based on similarity to the distribution of periodic components, and ARIMA (Autoregressive Integrated Moving Average). Some are based on the similarity of linear time series models such as, and some are based on the mutual information of two time series.

In estimating the cause of poor drying, it is considered effective to utilize these achievements and create a discrimination rule using a similarity scale. On the other hand, the similarity scale tends to function differently depending on the object to be handled, and evaluation is performed for each product / model and failure type for home appliances with various products, models, and failure types. It takes a huge amount of time and manpower to select an effective scale, and it is expected that system operation will be difficult. From such consideration, the discrimination rule is derived in the following frameworks (A) and (B).
(A) Clustering is performed using various similarity measures.
(B) Using the clustering result (classification type) as an explanatory variable, a discrimination rule is derived by machine learning.

The upper part of FIG. 9 shows the flow of deriving the discrimination rule according to such a framework. Each operation data of the training sample is classified and labeled by time series clustering, and a vector of label values is assigned to each operation data (S20). The discrimination rule RL is derived by machine learning using this vector as an explanatory variable and the presence or absence of a failure as an objective variable (S21). The above processing is performed for each cause of poor drying, and a discrimination rule RL for each cause is created.

In this embodiment, 16 typical types of similarity scales shown in FIG. 10 are used as similarity scales for time series clustering, centering on those having a track record in discriminating sensor time series in literature and the like. .. In addition, clustering is classified into, for example, 12 groups by hierarchical clustering. As a result, it becomes a 48-dimensional vector as an explanatory variable with 3 types of time series data and 16 types of clustering method. Clustering may be classified into, for example, 4 groups and 12 groups using a plurality of divisions. In this case, the explanatory variables are 96-dimensional vectors with 3 types of time series data, 16 types of clustering methods, and 2 types of divisions. Random forest and xgboost are used as the machine learning method for deriving the discrimination rule RL.

The lower part of FIG. 9 shows the flow of determining the cause of poor drying using the discrimination rule RL created as described above. Time-series data of operation data is acquired from the washer-dryer installed in each household, and clustering is performed to classify the time-series data into corresponding classification types for each of the above-mentioned 16 types of similarity scales (S22). Then, by comparing the classification results for each of the 16 types of similarity scales with the discrimination rule RL, it is determined whether or not the corresponding cause is the cause of the drying defect (S23). Such a discrimination process is performed for each cause of poor drying, and the cause of poor drying is estimated based on the discrimination result for each cause.

(2-2) Configuration of Drying Defect Cause Estimating System According to the Second Embodiment FIG. 11 is shown with the same reference numerals as those corresponding to FIG. 2 is the logical configuration of the drying defect cause estimation system 80 according to the second embodiment. Is shown. Since the hardware configuration of the dry defect cause estimation system 80 of the present embodiment is the same as that of the dry defect cause estimation system 1 of the first embodiment, the description thereof is omitted here.

The drying defect cause estimation system 80 includes the data acquisition unit 31 (FIG. 2) and the cause estimation unit 33 in which the configuration of the data acquisition unit 82 of the washing / drying machine 81 and the cause estimation unit 84 of the analysis server 83 is the first embodiment. The difference from FIG. 2 is different from the drying defect cause estimation system 1 of the first embodiment.

In practice, the data acquisition unit 82 of the washer / dryer 81 combines the acquired operation data DT1 of each sensor 13 for each sensor into time-series data for each sensor (hereinafter, this is referred to as operation data time-series data). ) It is transmitted to the analysis server 83 as DT3.

The analysis server 83 includes a cause estimation unit 84 including a plurality of determination units 85, and an estimation result output unit 86. The determination unit 85 is provided corresponding to each cause of drying failure such as clogging of the filter, clogging of the duct, and malfunction of the blower and the heater. Further, in the storage device 22 (FIG. 1) of the analysis server 83, the determination rule RL for each cause described above for FIG. 9 is stored in advance.

Then, each determination unit 85 describes the corresponding operation data time-series data DT3 among the various operation data time-series data DT3 transmitted from the washer-dryer 81, for example, with respect to FIG. Perform each clustering. Further, the determination unit 85 compares the clustering result of these clustering with the determination rule RL of the same cause, and determines whether the drying defect of the washer / dryer 81 is due to the corresponding cause based on the comparison result, and determines. The result is output to the estimation result output unit 86.

The estimation result output unit 86 estimates the cause of the drying failure in the corresponding washer / dryer 81 based on the determination result for each cause given by each determination unit 85, and displays the estimation result on the display device 24 (FIG. 1). Display etc.

(2-3) Learning flow of discrimination rule and flow of abnormality determination using discrimination rule FIG. 12 shows a configuration of a learning device 100 for learning the discrimination rule RL for each cause, and FIG. 13 shows the learning device. The flow of learning in 100 is shown.

The washer / dryer 90 shown in FIG. 12 is a washer / dryer used for creating a determination rule RL for each cause, and is a clock 91, a drive / power unit 92, a plurality of sensors 93, a control unit 94, and a data acquisition unit. 95 is provided. These clock 91, drive / power unit 92, each sensor 93, control unit 94, and data acquisition unit 95 correspond to the above-mentioned washing / drying machine 81 with respect to FIG. 11 (clock 11, drive / power unit 12, sensor 13, control). Since it has the same function and configuration as the unit 30 or the data acquisition unit 82), the description thereof is omitted here.

Similar to the data acquisition unit 82 of FIG. 11, the data acquisition unit 95 of the washer / dryer 90 is preset for a certain period before and after the control unit 94 outputs the command value COM10 to the drive / power unit 92. The operation data DT10 only in the time zone is acquired, the operation data time-series data DT11 combined with the acquired operation data DT10 is generated, and the generated operation data time-series data DT11 is output to the learning device 100.

The learning device 100 is a computer device provided with information processing resources such as a CPU, a memory, and a storage device (not shown). The learning device 100 may be used in combination with the analysis server 83 (FIG. 11). The cause estimation unit 101 of the learning device 100 is provided with a determination unit 102 and a determination rule RL corresponding to each cause of drying failure such as clogging of a filter, clogging of a duct, and failure of a blower or a heater.

Then, in each determination unit 102, among the operation data time series data DT11 given from the data acquisition unit 95 of the washing / drying machine 90, the operation data for learning necessary for learning the drying defect caused by the corresponding cause is used. Series data DT11 is given respectively (S30 in FIG. 13). For example, the determination unit 102 corresponding to the malfunction of the heater is given the operation data time series data DT11 generated from the operation data DT10 from the temperature sensor that measures the temperature of the air warmed by the heater.

Then, each determination unit 102 clusters the given operation data time series data DT11 with the above-mentioned 16 types of similarity scales with respect to FIG. At this time, each determination unit 102 is given an answer as to whether or not the cause of the drying defect is due to the corresponding cause, based on the user operation or the like for the learning device 100. Thus, each determination unit 102 creates the relationship information 104 between the classification type and the failure rate as shown in FIG. 9 based on this answer and the clustering result of the clustering of the 16 types of similarity scales described above. The created relationship information 104 is output to the corresponding machine learning unit 103 (S31 in FIG. 13).

The machine learning unit 103 learns the correlation between the clustering result of the above-mentioned 16 types of similarity scale clustering and the corresponding cause based on the relationship information 104 given by the corresponding determination unit 102 (FIG. 13). S32), the discrimination rule RL is updated as necessary based on the learning result (step S33 in FIG. 13).

Then, in the learning device 100, each time the operation data time series data DT11 is given from the washing / drying machine 90 for learning, the above-mentioned processes of steps S30 to S33 are repeated. As a result, learning of each discrimination rule RL corresponding to each cause is performed.

On the other hand, FIG. 14 shows the flow of the determination process executed on the analysis server 83 (FIG. 11) using each determination rule RL trained as described above. Each determination unit 85 (FIG. 11) of the analysis server 83 is given each operation data time series from the data acquisition unit 82 (FIG. 11) of the washer / dryer 81 (FIG. 11) via the network 4 (FIG. 1). Of the data DT3, operation data time-series data DT3 (FIG. 11) corresponding to the corresponding cause is given (S40).

Then, each determination unit 85 performs clustering of 16 types of similarity scales on the given operation data time series data DT3 (S41), and based on the clustering results of the clustering of these 16 types of similarity scales. It is determined whether or not the drying defect is due to the corresponding cause, and the determination result is output to the estimation result output unit 86 (FIG. 11) (S42).

(2-4) Effect of the present embodiment As described above, in the drying defect cause estimation system 80 of the present embodiment, the control unit 30 of the operation data DT1 output from each sensor 13 of the washing / drying machine 2 is used. The cause of failure of the washing / drying machine 2 is specified based on a certain period before and after the command value COM1 is output to the drive / power unit 12 and the time series data of the operation data DT1 in a predetermined specific time zone.

Therefore, according to the dry defect cause estimation system 80 of the present embodiment, the control unit 30 gives a command value COM1 to which drive / power unit 12 as in the dry defect cause estimation system 1 of the first embodiment. It is possible to specify whether or not the operation data DT1 shows an abnormal value when the above is output, and it is possible to identify the cause of the poor drying accordingly. Therefore, according to the drying defect cause estimation system 80, it is possible to efficiently estimate the cause of the defect in which a plurality of possible causes exist.

(3) Other Embodiments In the above-mentioned first embodiment, as described above with respect to FIG. 2, the case where the cause model 27 and the determination unit 32 for each cause are provided, respectively, has been described. The present invention is not limited to this, and as shown in FIG. 15 in which the corresponding portions corresponding to those in FIG. 2 are designated by the same reference numerals, one mixture in which the cause estimation unit 111 of the analysis server 110 is trained by mixing all the causes. It may be configured by the cause model 112 and the determination unit 113. Even in this way, the same effect as that of the first embodiment can be obtained. Similarly, in the second embodiment, the cause estimation unit of the analysis server may be configured by one discrimination rule and the discrimination unit in which all the discrimination rule RLs are mixed.

Further, in the first and second embodiments described above, the washer / dryer is a functional unit for estimating the cause of the drying failure of the washer / dryer 2,81 such as the cause estimation unit 33, 84 and the estimation result output unit 34, 86. Although the case where the analysis servers 3 and 83 are arranged separately from the 2,81 is described, the present invention is not limited to this, and these functional parts may be provided in the washer / dryer 2,81. ..

In this case, for example, the estimation result of the cause estimation unit is acquired from the washing / drying machine via a smartphone or tablet equipped with a short-range communication function such as Bluetooth (registered trademark), and the acquired estimation result is obtained in the network 4. It may be possible to transfer to the center of the manufacturer via Bluetooth.

Further, in the above-mentioned first and second embodiments, the case where the present invention is applied to the drying defect cause estimation system 1, 80 for estimating the cause of the drying defect of the washer / dryer 2,81 has been described. However, the present invention is not limited to this, and can be widely applied to various defect cause estimation devices for estimating the causes of defects other than drying defects. It can also be applied to a defect cause estimation device for estimating a defect cause of an electric power device other than the washer / dryers 2 and 81.

Further, in the second embodiment described above, a case where 16 types of similarity measures shown in FIG. 10 are applied as a similarity scale for time series clustering has been described, but the present invention is not limited to this. Some or all of them may use a similarity scale other than these 16 types of similarity scales, and the number of similarity scales may be other than 16 types.

The present invention can be widely applied to various defect cause estimation devices for estimating the defect cause of electric power equipment such as a washer / dryer.

1,80,90 ... Drying defect cause estimation system, 2,81 ... Washing and drying machine, 3,83,111 ... Analysis server, 12,42,92 ... Drive / power unit, 13,43,93 ... ... Sensor, 20 ... CPU, 27,112 ... Cause model, 30,44,94 ... Control unit, 31,45,82,95 ... Data acquisition unit, 32,51,85,102,113 ... Judgment unit, 33, 84, 111 ... Cause estimation unit, 34, 86 ... Estimated result output unit, 103 ... Machine learning unit, COM1, COM2 ... Command value, DT1, DT2, DT10 ... Operation data, DT3 , DT11 …… Operation data time series data, RL …… Discrimination rule.

Claims (8)

1. In the defect cause estimation device that estimates the defect cause of the target device,
   The target device is
   Each has a plurality of sensors for detecting the state of the target device.
   A storage unit that stores one or more models learned in advance,
   Using the sensor data in a specific data range that causes a state change of the target device among the sensor data output from the sensor, a cause estimation unit that estimates the cause of the defect of the target device based on the model. A defect cause estimation device characterized by being provided.
2. The target device is
   It has a control unit that controls the operation of the controlled object by giving the required command value to the corresponding controlled object at the required timing.
   The specific data range that causes the state change of the target device is a range of a predetermined period before and after the control unit gives the command value to the control target and / or a preset specific time zone. The defect cause estimation device according to claim 1.
3. The defect cause estimation device according to claim 1, wherein the model is a CNN (Convolutional Neural Network).
4. The defect cause estimation device according to claim 1, wherein the model is a discrimination rule for time-series clustering using a plurality of similarity measures.
5. In the defect cause estimation method executed by the defect cause estimation device for estimating the defect cause of the target device,
   The target device is
   Each has a plurality of sensors for detecting the state of the target device.
   The first step of storing one or more pre-trained models,
   A second step of estimating the cause of failure of the target device based on the model using the sensor data in a specific data range that causes a state change of the target device among the sensor data output from the sensor. A method for estimating the cause of a defect, which comprises.
6. The target device is
   It has a control unit that controls the operation of the controlled object by giving the required command value to the corresponding controlled object at the required timing.
   The specific data range that causes the state change of the target device is a range of a predetermined period before and after the control unit gives the command value to the control target and / or a preset specific time zone. The defect cause estimation method according to claim 5.
7. The defect cause estimation method according to claim 5, wherein the model is a CNN (Convolutional Neural Network).
8. The defect cause estimation method according to claim 5, wherein the model is a discrimination rule for time-series clustering using a plurality of similarity measures.

Images