Rail trolley system

Abstract

The present invention provides a rail trolley system. The rail trolley system (3) includes a traveling track (9) and a traveling trolley (7) that travels along the traveling track (9). The rail trolley system (3) includes a sensor (21) and a diagnostic device (17). The sensor (21) is fixedly arranged relative to the traveling track (9) to measure vibration or sound. The diagnostic device (17) diagnoses the state of the traveling trolley (7) that has passed through the measurement location corresponding to the installation position of the sensor (21) in the traveling track (9) based on the measurement data measured by the sensor (21).

Description

Technical Field

[0001] The present invention relates to a configuration for diagnosing the state of a trolley in a rail trolley system. Background Art

[0002] Conventionally, in a rail trolley system, the state of a trolley has been diagnosed by various methods in consideration of the possibility of failure. Patent Document 1 discloses a method for detecting deterioration of running wheels of a running trolley.

[0003] The running trolley disclosed in Patent Document 1 includes a drive wheel unit. The drive wheel unit includes components such as running wheels, a speed reducer, and a running motor. The running motor has an encoder. The rotational speed of the running wheels is detected by this encoder. A signal of the output torque and an encoder signal are taken out from the running motor, and the slip speed is obtained based on the difference between the encoder signal and the signal of a linear sensor provided in the running trolley. Deterioration of the running wheels is detected based on the fact that the output torque of the running motor and the slip speed pass through a specified abnormal region in the space (e.g., a two-dimensional plane) of the running torque and the slip speed.

[0004] Prior Art Documents

[0005] Patent Documents

[0006] Patent Document 1: Japanese Patent No. 6337528 Gazette Summary of the Invention

[0007] Problems to be Solved by the Invention

[0008] However, in the technology of the above Patent Document 1, in order to detect deterioration of the running wheels, it is necessary to provide an encoder and a linear sensor for all overhead running vehicles to be targeted, and there is room for improvement in this regard.

[0009] The present invention has been completed in view of the above circumstances, and an object thereof is to provide a rail trolley system capable of diagnosing the state of a trolley without providing a special sensor on the trolley side.

[0010] Means for Solving the Problems

[0011] The problems to be solved by the present invention are as described above. Hereinafter, the means for solving the problems and their effects will be described.

[0012] According to the viewpoint of the present invention, a rail trolley system configured as follows is provided. That is, the rail trolley system includes a track and a trolley that travels along the track. The above-mentioned rail trolley system includes a measurement unit, a diagnostic device, a feature quantity extraction unit, a learning control unit, a partial data extraction unit for learning, and a position sensor. The above-mentioned measurement unit is fixedly arranged relative to the position of the above-mentioned track and measures at least one of vibration and sound. The above-mentioned diagnostic device diagnoses the state of the trolley that has passed through the measurement location based on the measurement data measured by the above-mentioned measurement unit, and this measurement location corresponds to the installation position of the above-mentioned measurement unit on the above-mentioned track. The above-mentioned feature quantity extraction unit extracts the learning feature quantity included in the partial data for learning, which is the partial data corresponding to the data when the above-mentioned trolley passes through the above-mentioned measurement location in the measurement data measured by the above-mentioned measurement unit. The above-mentioned learning control unit forms a learning model related to the diagnosis of the state of the above-mentioned trolley using a set of learning feature quantities, that is, a learning data set, based on the measurement data measured by the above-mentioned measurement unit for a plurality of trolleys. The above-mentioned partial data extraction unit for learning extracts the above-mentioned partial data for learning from the measurement data measured by the above-mentioned measurement unit. The above-mentioned position sensor is fixedly arranged relative to the position of the above-mentioned track and detects the above-mentioned trolley passing through the above-mentioned measurement location. Based on the timing when the above-mentioned position sensor detects the above-mentioned trolley, the above-mentioned partial data is extracted from the above-mentioned measurement data.

[0013] Thereby, even if no special sensor is provided on the trolley side, the state of the trolley can be diagnosed. Since the measurement unit does not need to be provided on each traveling trolley, the cost can be reduced. The partial data for learning and the partial data for diagnosis can be obtained well.

[0014] In the above-mentioned rail trolley system, it is preferably configured as follows. That is, the above-mentioned feature quantity extraction unit extracts the diagnostic feature quantity included in the partial data for diagnosis, which is the partial data corresponding to the data when the above-mentioned trolley passes through the above-mentioned measurement location in the measurement data measured by the above-mentioned measurement unit. The above-mentioned diagnostic device calculates a machine learning evaluation value corresponding to the above-mentioned diagnostic feature quantity extracted by the above-mentioned feature quantity extraction unit based on the above-mentioned learning model learned by the above-mentioned learning control unit.

[0015] Thereby, the diagnosis of the state of the above-mentioned trolley can be performed well using the machine learning evaluation value.

[0016] In the above-mentioned rail trolley system, it is preferably configured as follows. That is, the rail trolley system includes a partial data extraction unit for diagnosis. The above-mentioned partial data extraction unit for diagnosis extracts the above-mentioned partial data for diagnosis from the measurement data measured by the above-mentioned measurement unit.

[0017] Thereby, the partial data for diagnosis can be obtained well.

[0018] In the above-mentioned rail trolley system, it is preferably configured as follows. That is, while shifting the above-mentioned measurement data variously in the time axis direction, the correlation between the reference data serving as a reference and the measurement data to be intercepted is obtained. In a state where the above-mentioned measurement data is shifted in the time axis direction such that the correlation becomes maximum, the above-mentioned partial data is intercepted from the above-mentioned measurement data in a time interval determined based on the above-mentioned reference data.

[0019] Thereby, partial data (partial data for learning and partial data for diagnosis) can be accurately obtained.

[0020] In the above-mentioned rail trolley system, it is preferably configured as follows. That is, the above-mentioned track is suspended from the ceiling of a building or a gantry installed on the ground. The above-mentioned trolley is an overhead transporter that travels along the above-mentioned track.

[0021] Since the overhead transporter is installed at a high place, generally, its maintenance work is difficult. However, according to this configuration, the state can be confirmed without removing the overhead transporter from the track, and the complexity of the maintenance work can be alleviated. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram showing the configuration of a rail trolley system according to an embodiment of the present invention.

[0023] Figure 2 It is a diagram showing the traveling track included in the rail trolley system.

[0024] Figure 3 It is a block diagram showing the configuration of a diagnostic device.

[0025] Figure 4 It is a flowchart of the processing performed in the rail trolley system to obtain a learning data set.

[0026] Figure 5 It is a flowchart of the processing for performing machine learning.

[0027] Figure 6 It is a flowchart of the processing performed in the rail trolley system by the state determination unit for determination.

[0028] Figure 7 It is a graph showing the relationship between the operation time of the traveling trolley and the evaluation value.

[0029] Figure 8 It is a graph for explaining the offset processing of the detection signal. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] Next, embodiments of the present invention will be described with reference to the drawings. Figure 1This is a schematic diagram showing the configuration of the rail vehicle system 3 according to an embodiment of the present invention. Figure 2 This is a diagram showing the traveling track 9 included in the rail vehicle system 3. Figure 3 This is a block diagram showing the configuration of the diagnostic device 17.

[0031] Figure 1 The rail vehicle system 3 shown is, for example, an automatic conveying system installed in a semiconductor manufacturing factory. The rail vehicle system 3 can convey a conveyance object 5 such as a FOUP. FOUP is an abbreviation for Front Opening Unified Pod.

[0032] The rail vehicle system 3 includes a plurality of traveling vehicles 7 and a traveling track (rail) 9.

[0033] The traveling vehicle 7 is a rail vehicle that travels along the traveling track 9. In the present embodiment, as the traveling vehicle 7, an overhead hoist transfer vehicle called an OHT is used. OHT is an abbreviation for Overhead Hoist Transfer. However, the traveling vehicle 7 is not limited thereto. In the rail vehicle system 3 of the present embodiment, a plurality of traveling vehicles 7 having substantially the same configuration travel simultaneously. Thereby, the conveyance efficiency can be improved.

[0034] As Figure 1 shown, each traveling vehicle 7 has a vehicle control unit 11. The vehicle control unit 11 is configured as a well-known computer having a CPU, ROM, RAM, HDD, etc.

[0035] The vehicle control unit 11 controls the transfer operation of the transfer mechanism such as a multi-joint robotic arm or a crane included in the traveling vehicle 7 for the conveyance object 5, and the automatic traveling etc. performed by the traveling mechanism (not shown) included in the traveling vehicle 7. The traveling vehicle 7 controlled by the vehicle control unit 11 travels along the traveling track 9 in a state where it is separated from other traveling vehicles 7 by an appropriate interval.

[0036] As Figure 2 shown, the traveling track 9 is installed in a building 13 such as a semiconductor manufacturing factory using the rail vehicle system 3. Specifically, the traveling track 9 is suspended from the ceiling of the building 13. In addition, the traveling track 9 can also be suspended from a pedestal provided on the ground. The traveling track 9 has, for example, a shape that guides the traveling path of the traveling vehicle 7 as Figure 1 shown. In addition, the shape of the traveling track 9 is not particularly limited.

[0037] The rail vehicle system 3 includes a learning system 1. As Figure 1 shown, the learning system 1 includes a measurement device 15, a diagnostic device 17, and a notification unit 45.

[0038] The measuring device 15 can measure at least one of vibration and sound generated as the traveling carriage 7 travels. In the present embodiment, the measuring device 15 is constituted by a sensor (measuring unit) 21. The sensor 21 detects the vibration or sound of the traveling track 9 when the traveling carriage 7 travels, and outputs a detection signal corresponding to the detection result of the sensor 21 to the diagnostic device 17. This detection signal corresponds to the measurement data measured by the sensor 21.

[0039] As described above, the sensor 21 of the present embodiment detects the vibration or sound of the traveling track 9. In the present embodiment, the sensor 21 is provided on the side of the traveling track 9 and not on the traveling carriage 7. In other words, the sensor 21 is provided in a position-fixed manner with respect to the traveling track 9. When the sensor 21 is a vibration sensor that detects vibration, this sensor is constituted by, for example, an acceleration sensor, an AE (Acoustic Emission) sensor, an ultrasonic sensor, or an impact pulse sensor, etc. When the sensor 21 is a sound sensor that detects sound, this sensor is constituted by, for example, a microphone, etc. Hereinafter, the position where the sensor 21 is arranged is sometimes referred to as the measurement location P1. As Figure 2 shown, the measurement location P1 is determined at an appropriate part on the traveling track 9. Although not shown, a special area called a diagnostic area is set near the measurement location P1.

[0040] A position sensor 25 is provided in the learning system 1. The position sensor 25 detects the traveling carriage 7 passing through the measurement location P1 on the traveling track 9. The position sensor 25 is provided in a position-fixed manner with respect to the traveling track 9.

[0041] The position sensor 25 is arranged on the traveling track 9 near the part corresponding to the measurement location P1. The position sensor 25 is constituted by a photoelectric sensor having a light transmitting and receiving part. When the light transmitted by the light transmitting and receiving part is detected to be reflected by a reflecting part of the traveling carriage 7 or the like through the light transmitting and receiving part, the position sensor 25 outputs a carriage detection signal to the diagnostic device 17. However, the constitution of the position sensor 25 is not limited to this.

[0042] The diagnostic device 17 can diagnose the state of the traveling carriage 7 that has passed through the measurement location P1 on the traveling track 9 based on the detection signal detected by the sensor 21. The diagnostic device 17 is electrically connected to the sensor 21 and the position sensor 25 respectively.

[0043] The diagnostic device 17 is constituted by a well-known computer having a CPU, a ROM, a RAM, an HDD, etc. The diagnostic device 17 performs various processes (including the diagnosis described later) by executing a program stored in a storage part constituted by the ROM, the RAM, the HDD, etc.

[0044] The notification unit 45 is composed of a display such as a liquid crystal monitor, a lamp, a buzzer, etc. The notification unit 45 is electrically connected to the diagnostic device 17. The notification unit 45 notifies an operator or the like of the result of the diagnosis of the state of the traveling carriage 7 by the diagnostic device 17 through display information and notification based on sound and / or light. Thus, the operator or the like can quickly know the state of the traveling carriage 7.

[0045] As Figure 1 and Figure 3 shown, the diagnostic device 17 has an AD conversion unit 37, a data extraction unit 31, a data conversion unit 33, and an analysis unit 35.

[0046] The AD conversion unit 37 converts the analog signal output by the sensor 21 into a digital signal. The AD conversion unit 37 outputs the converted digital signal to the data extraction unit 31.

[0047] The data extraction unit 31 extracts, from the detection signal output by the sensor 21, a part within a specified time range that includes the timing of vibrations or sounds generated in the traveling track 9 due to the passage of the traveling carriage 7. This time range is determined based on the timing of the carriage detection signal output by the position sensor 25. Hereinafter, the detection signal thus extracted may sometimes be referred to as a partial signal (partial data). Depending on the situation, the partial signal is used for learning or for diagnosis. Therefore, the data extraction unit 31 functions as a learning partial data extraction unit and a diagnostic partial data extraction unit.

[0048] The data conversion unit 33 converts the partial signal extracted by the data extraction unit 31 to obtain a frequency spectrum. The data conversion unit 33 has a feature quantity extraction unit 39.

[0049] The feature quantity extraction unit 39 obtains the frequency spectrum included in the partial signal extracted by the data extraction unit 31. Specifically, the feature quantity extraction unit 39 performs a discrete Fourier transform process on the partial signal output by the data extraction unit 31 as an object. Thus, a frequency spectrum representing the relationship between the frequency and the intensity of the signal vibrating at that frequency can be obtained. In the present embodiment, this frequency spectrum is used as a feature quantity (learning feature quantity or diagnostic feature quantity).

[0050] The analysis unit 35 forms a learning model by performing machine learning on a plurality of frequency spectra input from the data conversion unit 33. In the present embodiment, the learning model is a function that inputs a frequency spectrum and outputs an evaluation value. When forming the learning model, the analysis unit 35 uses this learning model to calculate an evaluation value for any traveling carriage 7 based on the frequency spectrum input from the data conversion unit 33. Since this evaluation value is based on machine learning, it can be called a machine learning evaluation value. The evaluation value obtained from the learning model is used for the diagnosis of the state of the traveling carriage 7.

[0051] The analysis unit 35 includes a machine learning unit (learning control unit) 41 and a state determination unit (diagnosis unit) 43.

[0052] The machine learning unit 41 uses the spectra of partial signals corresponding to a plurality of traveling bogies 7 as a set of learning data, and forms a learning model related to the diagnosis of the state of the traveling bogie 7. In the present embodiment, as the learning model, a known one-class SVM (One Class SVM) is used. SVM is the abbreviation of Support Vector Machine. In the present embodiment, outlier detection (in other words, anomaly detection) is performed through unsupervised learning of the one-class SVM.

[0053] Hereinafter, the construction of the learning model will be described in detail. Figure 4 And Figure 5 is a flowchart of the processing of the learning system 1 for the machine learning unit 41 to form a learning model.

[0054] First, the learning system 1 collects data for the machine learning unit 41 to form a learning model through Figure 4 the processing shown.

[0055] First, for the traveling bogie 7 traveling along the traveling track 9, a detection signal regarding vibration or sound generated due to traveling is acquired by the sensor 21 (step S101). Here, the traveling bogies 7 passing through the measurement point of the traveling track 9 are all normal traveling bogies 7. Hereinafter, the data obtained by measuring the normal traveling bogie 7 is sometimes referred to as normal data.

[0056] Next, the data extraction unit 31 of the diagnostic device 17 extracts a part of the signal from the detection signal acquired by the sensor 21 (step S102). Thus, the above partial signal is obtained. The time range for extracting the partial signal from the detection signal is from a specified time before the timing when the position sensor 25 detects that the traveling bogie 7 has passed through the measurement point P1 to a specified time after that timing.

[0057] Next, in the data conversion unit 33 of the diagnostic device 17, the feature quantity extraction unit 39 performs discrete Fourier transform processing on the partial signal from the data extraction unit 31 and extracts the spectrum (step S103). In the present embodiment, the feature quantity input to the learning model is this spectrum. Strictly speaking, the feature quantity is the intensity of the signal of each frequency component included in the spectrum. The spectrum is appropriately smoothed. There are various methods for smoothing processing, but for example, a method of performing convolution integration on a Gaussian filter can be cited.

[0058] Next, a storage unit (not shown) of the diagnostic device 17 stores the spectrum obtained by the feature quantity extraction unit 39 as learning data constituting the learning dataset (step S104).

[0059] The processes of steps S101 to S104 are repeated until the traveling carriage 7 has passed the measurement point P1 a sufficient number of times and a large number (e.g., several thousand) of spectra are obtained (step S105). In order to obtain spectra under the same conditions, the traveling speed of the traveling carriage 7 passing through the measurement point P1 for data collection is controlled to be constant at a prescribed data acquisition speed.

[0060] The measurement point P1 set on the traveling track 9 allows not only one traveling carriage 7 to pass through, but also multiple traveling carriages 7 to pass through. In the present embodiment, if, for example, several hundred spectra are obtained for one normal traveling carriage 7, the operation of collecting spectra of the same degree is performed by replacing it with another normal traveling carriage 7 as an individual (step S106).

[0061] Through the above processing, a learning dataset can be obtained. In the present embodiment, the learning dataset is substantially a collection of a large number of spectra. This learning dataset is stored in the storage unit of the diagnostic device 17. In addition, the learning dataset stored in this storage unit can be appropriately changed (updated).

[0062] Due to the carriage replacement operation in step S106, the learning dataset obtained through Figure 4 the processing contains spectra obtained from multiple traveling carriages 7 (different individuals). By performing machine learning using such a learning dataset, it is possible to suppress overlearning of the characteristics unique to the individual of the traveling carriage 7. Regarding the number of traveling carriages 7 from which spectra are collected, as long as it is two or more, it is arbitrary, but it is preferably set to a relatively large number such that the individual differences of the traveling carriages 7 become inconspicuous.

[0063] The learning dataset does not contain spectra obtained from abnormal traveling carriages 7, and all spectra are obtained from normal traveling carriages 7.

[0064] Next, the learning system 1 uses the obtained learning dataset to perform machine learning based on the machine learning unit 41 through Figure 5 the processing shown. Figure 5 The processing in

[0065] corresponds to the training phase of machine learning. First, in the analysis unit 35 of the diagnostic device 17, the machine learning unit 41 appropriately initializes the parameters of the discriminant formula for discriminating the abnormality of the traveling carriage 7 (step S201). In the present embodiment, the discriminant formula is expressed as follows.

[0066] [Equation 1]

[0067]

[0068]

[0069] In the above equation, f(x) is the discriminant, which is equivalent to the learning model of the present embodiment. The input x to the discriminant f is an N-dimensional vector representing the input spectrum. In the present embodiment, when the spectrum is represented by the signal intensities of multiple frequency components, N refers to the number of such frequency components. The output of the discriminant f is a scalar value called the evaluation value, and the details will be described later.

[0070] K(xi,x) is a known kernel function of one-class SVM. The kernel function is a mapping function to a high-dimensional space such that the more the spectrum becomes an outlier, the closer it is to the origin. In one-class SVM, in the mapping to the high-dimensional space, a hyperplane is determined where the distance from the origin is maximized. This hyperplane becomes the criterion for determining outliers. In the present embodiment, a Gaussian kernel is used as the kernel function, but the kernel function is not limited to this.

[0071] αi, xi, and ρ are parameters to be the object of machine learning. In the present embodiment, forming the learning model is equivalent to obtaining the optimal values of αi, xi, and ρ to determine the discriminant. In step S201, these parameters are initialized. The values used for initialization (initial values of the parameters) can be set to random values, for example. The other parameter, σ, is called a hyperparameter and is appropriately determined by the designer.

[0072] Next, the machine learning unit 41 extracts one of the spectra included in the learning dataset, substitutes it into the discriminant f in the form of an N-dimensional vector x, and obtains the value f(x) (step S202). Then, αi, xi, and ρ are adjusted so that the obtained value f(x) approaches 0 (step S203). The processes of step S202 and step S203 are repeated for all the spectra in the learning dataset (step S204). Thus, one round of learning is completed. A round means performing one learning on the learning data included in the learning dataset.

[0073] The processes of step S202 to step S204 are further repeated an appropriate number of times (step S205). Thus, multi-round learning is achieved. By repeatedly performing learning an appropriate number of times on one learning dataset, a learning model with good performance is obtained.

[0074] By Figure 5The values of the parameters αi, xi, and ρ are obtained through the processing, and these parameter values are stored in the storage unit of the diagnostic device 17. The discriminant formula f substituted with the values of the parameters αi, xi, and ρ represents the learned model. Therefore, it can also be considered that the obtained values of the parameters αi, xi, and ρ themselves are substantially the learned model.

[0075] In the outlier detection using one-class SVM, if the value of f(x) is 0 or positive, it means that x is not an outlier. If the value of f(x) is negative, it means that x is an outlier, and the smaller the value of f(x), the greater the degree of outlier of the value. The state discrimination unit 43 of the present embodiment diagnoses the normality / abnormality of the traveling carriage 7 by using the relationship that the spectrum when the traveling carriage 7 generates an abnormality is an outlier with respect to the spectrum in the normal state.

[0076] In the actual diagnosis, the following Figure 6 processing is performed. Figure 6 The processing is equivalent to the inference stage of machine learning.

[0077] First, vibrations or sounds caused by the traveling of the traveling carriage 7 are acquired by the sensor 21 (step S301). At this time, the traveling carriage 7 is controlled to pass through the measurement point P1 at the same speed as the above data acquisition speed. A partial signal is intercepted from the detection signal (step S302). Hereinafter, this partial signal is sometimes referred to as the diagnostic partial signal. Further, a spectrum is extracted from the diagnostic partial signal (step S303).

[0078] The processing of steps S301 to S303 is substantially the same as Figure 4 steps S101 to S103. However, in Figure 6 this case, it is not clear whether the traveling carriage 7 passing through the measurement point P1 generates an abnormality.

[0079] Next, the state discrimination unit 43 inputs the spectrum obtained in step S303 into the discriminant formula and calculates an evaluation value (step S304). At this time, as the parameters αi, xi, and ρ of the discriminant formula, the parameters obtained through Figure 5 the processing are used. Therefore, the discriminant formula used in step S304 is synonymous with the learned learning model.

[0080] Next, the state discrimination unit 43 determines whether the calculated evaluation value is less than the threshold value v1 (step S305). In Figure 7 the graph shows the change in the evaluation value when the same traveling carriage 7 continues to operate. After completing Figure 5 the training stage in the learning model (discriminant formula), the durability test of the traveling carriage 7 is actually performed, etc., and thus this graph can be obtained.

[0081] In the present embodiment, as shown in Figure 7 the curve graph, the threshold value v1 is set to a value assuming that the traveling bogie 7 is about to fail.

[0082] Specifically, in the curve graph of Figure 7 , the traveling bogie 7 fails at the timing of t2 and becomes inoperable. In this case, it is preferable to generate a certain warning at a timing slightly before t2, for example, t1. Therefore, the threshold value v1, which is the discrimination boundary between normal and abnormal, is determined as a value with a certain margin preset with respect to the evaluation value corresponding to a state where an obvious failure is confirmed in the traveling bogie 7 (hereinafter referred to as the failure evaluation value v2). The threshold value v1 and the failure evaluation value v2 are negative values. The failure evaluation value v2 can be determined with reference to the results of the above-mentioned durability test, etc. The threshold value v1 is larger than the failure evaluation value v2 by the amount of the above-mentioned margin. The size of this margin is appropriately determined in consideration of the importance of the continuous operation of the rail trolley system 3 and the component replacement cost, etc.

[0083] In the determination of step S305, when the evaluation value is equal to or less than the threshold value v1, the state discrimination unit 43 discriminates that the state of the traveling bogie 7 is abnormal (step S306). In this case, the state discrimination unit 43 sends the discrimination result (the state of the traveling bogie 7 is abnormal) to the notification unit 45 (step S307). As a result, in the notification unit 45, for example, one or both of "the state of the traveling bogie 7 is abnormal" and "it is necessary to prepare a replacement traveling bogie or component for the traveling bogie 7" are notified. The length of the remaining operable time is the time length until the timing when it is predicted that the evaluation value becomes equal to or less than the failure evaluation value v2. For example, it can be inferred by Figure 7 the curve graph and the current evaluation value.

[0084] In the determination of step S305, when the evaluation value is equal to or greater than the threshold value v1, the state discrimination unit 43 determines that the state of the traveling bogie 7 is normal (step S308).

[0085] Regardless of whether the discrimination result is normal or abnormal, the process returns to step S301, and diagnosis is performed again based on the passage of the traveling bogie 7.

[0086] In addition, the notification of the notification unit 45 preferably includes information indicating the meaning of the need to prepare components for the traveling bogie 7 or a replacement traveling bogie. In this case, the operator, etc. can be urged to make appropriate preparations.

[0087] As described above, the rail vehicle system 3 of the present embodiment includes a traveling track 9 and a traveling vehicle 7 that travels along the traveling track 9. The rail vehicle system 3 includes a sensor 21 and a diagnostic device 17. The sensor 21 is fixedly arranged relative to the traveling track 9 and measures vibration or sound. The diagnostic device 17 diagnoses the state of the traveling vehicle 7 that has passed through the measurement point P1 corresponding to the installation position of the sensor 21 in the traveling track 9 based on the measurement data measured by the sensor 21.

[0088] Thereby, even without providing a special sensor on the traveling vehicle 7 side, the state of the traveling vehicle 7 can be diagnosed. Since the sensor 21 can be not provided on each traveling vehicle 7, the cost can be reduced.

[0089] The rail vehicle system 3 of the present embodiment includes a feature quantity extraction unit 39 and a machine learning unit 41. The feature quantity extraction unit 39 extracts the frequency spectrum included in the partial signal, that is, the learning partial signal, which corresponds to the detection signal when the traveling vehicle 7 passes through the measurement point P1 in the detection signal detected by the sensor 21. The machine learning unit 41 forms a learning model related to the state diagnosis of the traveling vehicle 7 using a learning data set, which is a set of frequency spectra of detection signals measured by the sensor 21 for a plurality of traveling vehicles 7.

[0090] Thereby, a learning model can be obtained well.

[0091] The rail vehicle system 3 of the present embodiment includes a data truncation unit 31. The data truncation unit 31 truncates the learning partial signal from the detection signal detected by the sensor 21.

[0092] Thereby, the learning partial signal can be obtained well.

[0093] In the rail vehicle system 3 of the present embodiment, the feature quantity extraction unit 39 extracts the frequency spectrum included in the partial signal, that is, the diagnostic partial signal, which corresponds to the detection signal when the traveling vehicle 7 passes through the measurement point P1 in the detection signal detected by the sensor 21. The diagnostic device 17 calculates an evaluation value corresponding to the frequency spectrum extracted by the feature quantity extraction unit 39 based on the learning model learned by the machine learning unit 41.

[0094] Thereby, the state of the traveling vehicle 7 can be diagnosed well using the evaluation value.

[0095] The rail vehicle system 3 of the present embodiment includes a data truncation unit 31. The data truncation unit 31 truncates the diagnostic partial signal from the detection signal detected by the sensor 21.

[0096] Thereby, the diagnostic partial signal can be obtained well.

[0097] The rail trolley system 3 of the present embodiment includes a position sensor 25. The position sensor 25 is fixedly arranged relative to the traveling track 9 and detects the traveling trolley 7 passing through the measurement point P1. The data intercepting unit 31 intercepts partial signals (learning partial signals and diagnostic partial signals) based on the timing when the position sensor 25 detects the traveling trolley 7.

[0098] Thereby, partial signals can be obtained better.

[0099] In the rail trolley system 3 of the present embodiment, the traveling track 9 is suspended from the ceiling of the building 13. The traveling trolley 7 is an overhead transporter that travels along the traveling track 9.

[0100] Since the overhead transporter is arranged at a high place, generally, its maintenance work is complicated. However, according to the present embodiment, the state of the traveling trolley 7 as an overhead transporter can be confirmed without removing it from the traveling track 9, and the complexity of the maintenance work can be alleviated.

[0101] Next, a modification related to the interception performed by the data interception unit 31 will be described.

[0102] In the above embodiment, based on the timing when the position sensor 25 outputs the trolley detection signal, the range for intercepting partial signals from the detection signal is determined. However, in this configuration, for example, due to individual differences in the mounting positions of the wheels of the traveling trolley 7, when the traveling trolley 7 passes through the measurement point P1, there is a deviation in the timing of vibration or sound generated on the traveling track 9, which may reduce the accuracy of the discrimination result. Therefore, in this modification, correction for adjusting the detection signal to be intercepted in the time axis direction is performed.

[0103] In the learning system 1 of the modification, before collecting data for the learning dataset, a reference signal (reference data) serving as a reference for determining the interception section of the data by the data interception unit 31 is prestored in the storage unit of the diagnostic device 17. The traveling trolley 7 is made to pass through the measurement point P1, and based on the output result of the AD conversion unit 37 at this time, the reference signal can be obtained. As the traveling trolley 7 at this time, a traveling trolley 7 in a normal state is used, and the traveling trolley 7 passes through the measurement point P1 at the above data acquisition speed.

[0104] After that, the Figure 4 processing starts. In the Figure 8 graph, the relationship between the waveform of the detection signal obtained in step S101 and the waveform of the above reference signal is illustrated in the case where the sensor 21 is a vibration sensor. However, in the Figure 8In [the description], the detection signal is simplified for convenience. The data truncation unit 31 calculates the correlation with the reference signal 51 while shifting the obtained detection signal 53 in various directions on the time axis, thereby searching for the shift direction and amount with the maximum correlation. The data truncation unit 31 shifts the detection signal 53 in the time axis direction so that the correlation of the detection signal 53 with respect to the reference signal 51 becomes maximum.

[0105] In this modification, the data truncation unit 31 truncates the time interval of the detection signal, and determines it based on the reference signal instead of the timing when the position sensor 25 outputs the bogie detection signal. The data truncation unit 31 truncates the detection signal in this time interval to obtain a partial signal.

[0106] The shift of the detection signal is not only performed in the Figure 4 processing, but also performed in the Figure 6 processing in the same way. In this modification, by processing to eliminate the deviation of the timing of the waveform acquired by the sensor 21, the diagnostic accuracy of the learning model can be improved.

[0107] As described above, in the rail bogie system 3 of this modification, the data truncation unit 31 calculates the correlation between the reference signal 51 serving as the reference and the detection signal 53 to be truncated while shifting the detection signal 53 in various directions on the time axis. The data truncation unit 31 truncates a partial signal from the detection signal 53 in the time interval determined based on the reference signal 51 in a state where the detection signal 53 is shifted in the time axis direction so that the correlation becomes maximum.

[0108] Thereby, a partial signal can be obtained in a form that eliminates the deviation of the timing.

[0109] The preferred embodiments of the present invention have been described above, but the above configurations can be changed as follows.

[0110] In the above embodiment, the machine learning unit 41 and the state determination unit 43 are provided in one diagnostic device 17, but they can also be provided in different devices. For example, only the Figure 3 machine learning unit 41 can be taken out and used as a device different from the diagnostic device 17. Hereinafter, a computer having the machine learning unit 41 and different from the diagnostic device 17 is called a training device. In this example, the place where the learning model is formed is different from the place where the state of the traveling bogie 7 is diagnosed. The learning data set obtained by data collection in the diagnostic device 17 is provided to the training device. The data of the learning model formed by the training device (substantially the parameters of the above discriminant) is provided to the diagnostic device 17. The provision of data between devices can be performed, for example, by known communication.

[0111] The state determination unit 43 can also determine that the state of the traveling carriage 7 is approaching abnormality when the state where the decrease amount of the evaluation value per unit time is large continues continuously (for example, at the t3 timing of Figure 7 ). Similar to the case where the evaluation value is less than the threshold value v1, the notification unit 45 can be used to notify the determination result.

[0112] As long as the above teachings are considered, the present invention can obviously obtain many modification methods and deformation methods. Therefore, it should be understood that the present invention can be implemented by methods other than those described in this specification within the scope of the appended patent claims.

[0113] Explanation of symbols

[0114] 3: Rail trolley system; 7: Traveling carriage (trolley); 9: Traveling track (track); 13: Building; 17: Diagnostic device; 21: Sensor (measurement unit); 25: Position sensor; 31: Data extraction unit (learning partial data extraction unit, diagnostic partial data extraction unit); 39: Feature quantity extraction unit; 41: Machine learning unit (learning control unit); 51: Reference signal (reference data); 53: Detection signal (measurement data).

Claims

1. A rail trolley system includes a track and a trolley traveling along the track, characterized in that, Comprising: A measurement unit, which is fixedly arranged relative to the above-mentioned track position and measures at least one of vibration and sound; A diagnostic device, which diagnoses the state of the trolley passing through the measurement location based on the measurement data measured by the above-mentioned measurement unit, and this measurement location corresponds to the installation position of the above-mentioned measurement unit on the above-mentioned track; A feature quantity extraction unit, which uses a part of the measurement data measured by the above-mentioned measurement unit when the above-mentioned trolley passes through the above-mentioned measurement location as learning partial data, and extracts the learning feature quantity included in this learning partial data; A learning control unit, which forms a learning model related to the diagnosis of the state of the above-mentioned trolley by using a set of learning feature quantities, namely a learning data set, based on the measurement data measured by the above-mentioned measurement unit for multiple trolleys; A learning partial data intercepting unit, which intercepts the above-mentioned learning partial data from the measurement data measured by the above-mentioned measurement unit; and A position sensor, which is fixedly arranged relative to the above-mentioned track position and detects the above-mentioned trolley passing through the above-mentioned measurement location, At the timing when the above-mentioned position sensor detects the above-mentioned trolley, the above-mentioned learning partial data is intercepted from the above-mentioned measurement data.

2. The rail trolley system according to claim 1, wherein: The above-mentioned feature quantity extraction unit uses the above-mentioned partial data as diagnostic partial data, and extracts the diagnostic feature quantity included in this diagnostic partial data, The above-mentioned diagnostic device calculates a machine learning evaluation value corresponding to the above-mentioned diagnostic feature quantity extracted by the above-mentioned feature quantity extraction unit based on the above-mentioned learning model learned by the above-mentioned learning control unit.

3. The rail trolley system according to claim 2, wherein: It further comprises a diagnostic partial data intercepting unit, which intercepts the above-mentioned diagnostic partial data from the measurement data measured by the above-mentioned measurement unit.

4. The rail trolley system according to claim 2 or 3, wherein: While shifting the above-mentioned measurement data in the time axis direction in various ways, the correlation between the reference data serving as the reference and the measurement data to be intercepted is obtained, In the state where the above-mentioned measurement data is shifted in the time axis direction so that the above-mentioned correlation becomes the maximum, in the time interval determined based on the above-mentioned reference data, the above-mentioned learning partial data or the above-mentioned diagnostic partial data is intercepted from the above-mentioned measurement data.

5. The rail trolley system according to any one of claims 1 to 3, wherein: The above-mentioned track is suspended from the ceiling of a building or a gantry installed on the ground, The above-mentioned trolley is an overhead transporter running along the above-mentioned track.

6. The rail trolley system according to claim 4, wherein: The above-mentioned track is suspended from the ceiling of a building or a gantry installed on the ground, The above-mentioned trolley is an overhead transporter running along the above-mentioned track.

Images