Fastening abnormality detection device and vehicle

Abstract

The present invention provides a fastening abnormality detection device that can determine with high accuracy whether or not a fastening loosen in a fastening structure. [Solution] The fastening abnormality determination device (ECU 500) includes a storage device 520 that stores a trained model. The trained model is machine-trained to output output information indicating whether or not fastening loosening has occurred in the fastening structure when predetermined input information is received. The input information includes waveform data showing the waveform of the current flowing through the fastening structure. The fastening abnormality determination device is configured to acquire waveform data using a current sensor, input the acquired waveform data into the trained model, and determine whether or not fastening loosening has occurred in the fastening structure based on the output information output from the trained model.

Description

Technical Field

[0004] ,

[0006] , , , , , ,

[0005] , , , , , ,

[0001] The present disclosure relates to a fastening abnormality determination device and a vehicle.

Background Art

[0002] Japanese Unexamined Patent Application Publication No. 2019-132608 (Patent Document 1) discloses a technique for detecting loosening of a fastening part when the number of times the amount of change in the electrical resistance of the fastening part periodically acquired exceeds a first threshold value is greater than a second threshold value. The fastening part is a part where a bus bar is fastened to an external terminal of a battery. The electrical resistance of the fastening part is calculated based on the current value of the fastening part (battery current value).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the above technique, based on the number of times the amount of change in the electrical resistance of the fastening part exceeds a threshold value, it is determined whether loosening has occurred in the fastening part (fastening structure). However, in a form in which the battery current value is controlled so as to follow a fluctuating target value, depending on the way the target value changes, the number of times the amount of change in the electrical resistance of the fastening part exceeds the threshold value may increase even if no loosening has occurred in the fastening part. In the above technique, misjudgment is likely to occur under specific circumstances.

[0005] The present disclosure has been made to solve the above problems, and an object thereof is to provide a fastening abnormality determination device and a vehicle capable of accurately determining whether loosening has occurred in a fastening structure.

Means for Solving the Problems

[0006] According to one embodiment of the present disclosure, the following fastening anomaly determination device is provided. The fastening anomaly determination device is configured to determine whether or not fastening loosening has occurred in a fastening structure composed of multiple fastened parts fastened together with a fastening part. The fastening anomaly determination device includes a storage device for storing a trained model. The trained model is machine-trained to output output information indicating whether or not fastening loosening has occurred in the fastening structure when predetermined input information is input. The input information includes waveform data showing the waveform of the current flowing through the fastening structure. The fastening anomaly determination device is configured to acquire waveform data using a current sensor, input the acquired waveform data into the trained model, and determine whether or not fastening loosening has occurred in the fastening structure based on the output information output from the trained model. [Effects of the Invention]

[0007] According to this disclosure, it becomes possible to provide a fastening abnormality detection device and a vehicle that can determine with high accuracy whether or not fastening loosening has occurred in a fastening structure. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows the schematic configuration of a vehicle according to an embodiment of the present disclosure. [Figure 2] This diagram shows the configuration of the fastening abnormality detection device according to this embodiment. [Figure 3] This is a time chart illustrating the training images used in the first learning stage. [Figure 4] This is a time chart illustrating the training images used in the second learning stage. [Figure 5] This flowchart shows the fastening loosening detection method according to this embodiment. [Figure 6] This flowchart shows a modified version of the processing flow shown in Figure 5. [Modes for carrying out the invention]

[0009] Embodiments of the present disclosure will be described in detail with reference to the drawings. In the drawings, the same or corresponding parts are denoted by the same reference numerals and their descriptions will not be repeated. In each drawing, the orientation of the three mutually orthogonal axes (X-axis, Y-axis, and Z-axis) is indicated by a "+" sign in the direction pointed to by the arrow, and a "-" sign in the opposite direction.

[0010] Figure 1 shows the configuration of a vehicle according to this embodiment. In Figure 1, the -X side corresponds to the direction of travel of the vehicle, and the -Z side corresponds to the vertical direction (direction of gravity). Referring to Figure 1, the vehicle 1000 is equipped with a battery pack 100. The battery pack 100 is fixed, for example, under the floor of the vehicle 1000. However, the mounting configuration of the battery pack 100 is arbitrary.

[0011] The vehicle 1000 further comprises a drive unit 20 for driving the vehicle 1000, an inlet 410 and charger 420 (onboard charger) used for charging the battery pack 100, and an ECU (Electronic Control Unit) 500.

[0012] The drive unit 20 includes a PCU (Power Control Unit) 21, an MG (Motor Generator) 22, and an engine 23. The vehicle 1000 is configured to run using the power output from the battery pack 100. The vehicle 1000 is, for example, a PHEV (Plug-in Hybrid Electric Vehicle). However, the vehicle 1000 may also be another electric vehicle (xEV), such as a BEV (Battery Electric Vehicle).

[0013] The PCU21 includes, for example, an inverter. The MG22 functions as a drive motor and rotates the drive wheels 24 of the vehicle 1000. The MG22 drives the vehicle 1000 using power output from the batteries in the battery pack 100. Specifically, the PCU21 drives the MG22 using power supplied from the battery pack 100. This puts the MG22 into a powered state. In the powered state, the MG22 converts power into torque. The torque is transmitted to the drive wheels 24. The MG22 also enters a regenerative state, for example when the vehicle 1000 is decelerating, and charges each battery in the battery pack 100 through regenerative power generation.

[0014] Engine 23 functions as an internal combustion engine, using the combustion energy of fuel to drive the vehicle 1000. Specifically, engine 23 generates power from the combustion energy of fuel supplied from a fuel tank (not shown). The generated power is transmitted to the drive wheels 24. An exhaust pipe 23a is connected to engine 23 and discharges the exhaust from engine 23 to the outside of the vehicle.

[0015] The battery pack 100 comprises a plurality of cells 10 (energy storage cells), each of which functions as a secondary battery. In this embodiment, liquid lithium-ion batteries are used as cells 10. However, cells 10 are not limited to lithium-ion batteries and may be other secondary batteries such as nickel-metal hydride batteries or sodium-ion batteries. The type of secondary battery is not limited to liquid secondary batteries and may be an all-solid-state secondary battery.

[0016] Each of the multiple cells 10 is provided with a negative terminal 11 and a positive terminal 12. A battery stack is formed when the multiple cells 10 are stacked and constrained, for example, in the X direction. The battery stack is an energy storage module in which multiple electrically connected cells 10 are modularized. Specifically, a spacer 15 is provided between adjacent cells 10 in the X direction. The spacer 15 may also function as a cooler to cool two cells 10 located on either side of it (+X side and -X side). The multiple cells 10 are electrically connected in series, with their orientations reversed one by one in the X direction. The negative terminal 11 of one cell 10 and the positive terminal 12 of the adjacent cell 10 are electrically connected via a conductive member 13 (e.g., a busbar). In detail, the terminals of each cell are fastened to the conductive member 13, as will be explained below.

[0017] Cell 10 comprises a metal case C1. The case C1 houses the electrode body C2 and electrolyte that constitute the lithium-ion battery. The positive electrode terminal 12 comprises a current collector terminal E1, a gasket E2, and a base E3. The current collector terminal E1 is housed within the case C1. The electrode body C2 includes a laminate of a plurality of positive electrode sheets and a plurality of negative electrode sheets. This laminate is formed by alternately stacking positive electrode sheets and negative electrode sheets. The positive electrode sheet includes, for example, a positive electrode current collector and a positive electrode active material layer. The negative electrode sheet includes, for example, a negative electrode current collector and a negative electrode active material layer. Each electrode sheet may be formed by coating the surface of a metal foil that serves as a current collector with an active material. A separator may be placed between the positive electrode sheet and the negative electrode sheet. Inside the case C1, the current collector terminal E1 of the positive electrode terminal 12 is electrically connected to the positive electrode sheet of the electrode body C2.

[0018] The gasket E2 is positioned between the case C1 and the pedestal E3. The pedestal E3 may be formed of a metal (for example, an alloy containing at least one of aluminum, iron, and copper). The current collecting terminal E1 may protrude outside the case C1 in a manner that penetrates the gasket E2 and the pedestal E3 from within the case C1. Then, the protruding portion of the current collecting terminal E1 may be fixed to the pedestal E3 by caulking on the upper surface (+Z side surface) of the pedestal E3.

[0019] A nut E4 and a bolt E5 are provided on the positive electrode terminal 12. The nut E4 and the bolt E5 are fastening parts configured to fasten the pedestal E3 of the positive electrode terminal 12 and the conductor member 13. The conductor member 13 may be a metal plate. The bolt E5 includes a head embedded in the gasket E2 and a threaded portion having a thread on its outer surface. In the cell 10, the thread formed on the inner surface of the nut E4 and the thread of the threaded portion of the bolt E5 are screwed together, whereby the pedestal E3 of the positive electrode terminal 12 and the conductor member 13 are fastened. A fastening structure (the fastening portion of the cell 10) is formed by fastening a plurality of fastened parts (the pedestal E3 and the conductor member 13) with the fastening parts (the nut E4 and the bolt E5).

[0020] [[ID=८]]The negative electrode terminal 11 basically has the same configuration as the positive electrode terminal 12. However, since the polarities of the negative electrode terminal 11 and the positive electrode terminal 12 are opposite, an appropriate material is selected for each terminal. Inside the case C1, the current collecting terminal of the negative electrode terminal 11 is electrically connected to the negative electrode sheet of the electrode body C2. The potentials of the negative electrode sheet and the positive electrode sheet of the electrode body C2 are output to the negative electrode terminal 11 and the positive electrode terminal 12 (external terminals), respectively.

[0021] As described above, in the battery stack within the battery pack 100, a plurality of fastening portions (fastening portions for each terminal) are formed. The number of cells 10 can be arbitrarily set according to required specifications (e.g., output power). The battery stack may include only the same type of cells or may include different types of cells. The electrode body C2 is not limited to a laminate in which a plurality of electrode sheets are laminated in one direction, and may be a wound body (e.g., a wound body in which a laminate of alternately arranged positive electrode sheets and negative electrode sheets is wound). A gas discharge valve may be provided in the case C1.

[0022] FIG. 2 is a diagram showing the configuration of the ECU 500. Referring to FIG. 2, the ECU 500 includes a processor 510 and a storage device 520. The storage device 520 is configured to be able to store the stored information. In the ECU 500, various controls are executed by the processor 510 executing the program stored in the storage device. In addition to the program, the storage device 520 further stores various information used in the program. Specifically, the storage device 520 stores the first learned model and the second learned model respectively generated by the first learning and the second learning described below.

[0023] In this embodiment, an unlearned neural network for the first learning and an unlearned neural network for the second learning are prepared separately. For example, a learned neural network (learned model) can be obtained by performing supervised machine learning on the unlearned neural network. In this embodiment, the unlearned neural network is a general-purpose machine learning algorithm. The learned model functions as a model for determining loosening of each fastening portion in the battery stack.

[0024] A neural network comprises an input layer Nx, a hidden layer Ny, and an output layer Nz. The input layer Nx receives a training image as input. The input layer Nx may contain a number of nodes corresponding to the number of pixels in the training image (N nodes). The output layer Nz has, for example, two nodes. Of the two nodes, one node outputs the probability (low: 0, high: 1) that the waveform corresponds to one with slack, and the other node outputs the probability (low: 0, high: 1) that the waveform does not have slack. However, the number of nodes in the output layer Nz is not limited to this, and it may be as simple as one.

[0025] In this embodiment, supervised machine learning was performed on an untrained neural network using training images, which represent the waveform of the current flowing through cell 10 (hereinafter referred to as the "battery current waveform") in a predetermined graph format and image format, as training data, along with ground truth data and additional information. The training image is data that shows the numerical value for each pixel (i.e., pixel value) in an image in which the current waveform is drawn in a predetermined area of ​​pixels. Hereinafter, the area in which the current waveform is drawn will also be referred to as the "image area". The number of pixels in the image area can be set arbitrarily. In one example of an image area, the number of vertical pixels is approximately 60 and the number of horizontal pixels is approximately 200. Each pixel value in the image area takes on either a value of 0 (white) or 1 (black). The ground truth data is data that indicates either no slack or slack occurring. In this embodiment, the target value of the current flowing through cell 10 (hereinafter simply referred to as the "current target value") is adopted as additional information.

[0026] Figure 3 is a time chart illustrating the training images used in the first learning stage. In Figure 3, lines L1 and L2 show data for normal and abnormal conditions, respectively. Under normal conditions, there is no loosening in any of the fastening points in the battery stack. Under abnormal conditions, there is loosening in one of the fastening points in the battery stack. Lines L1 and L2 each show the progression of the battery current (current of cell 10). Line L10 shows the progression of the current target value. "t" in the time chart represents timing.

[0027] In the example shown in Figure 3, at t1, the current target value decreases from the first target value (hereinafter referred to as "I1") to a second target value (hereinafter referred to as "I2") which is smaller than I1. I1 corresponds to the current target value before the decrease, and I2 corresponds to the current target value after the decrease. The period from t1 to t2 corresponds to the current convergence period. The start timing of the current convergence period is the timing when the current target value is changed. The length of the current convergence period corresponds to the time it takes from the change in the current target value (increase or decrease) until the battery current converges. The length of the current convergence period is determined experimentally in advance for each of the increases and decreases in the current target value, for example, and stored in the memory device 520. In one example, the length of the current convergence period for the decrease in the current target value was 2 seconds.

[0028] As shown by line L1, under normal conditions, the battery current converges to within ±10% of the target current value by the time the current convergence period (period t1-t2) has elapsed. Thus, when the target current value is changed under normal conditions, the battery current changes to approach the target current value and converges to a range close to the changed target current value. In contrast, under abnormal conditions, as shown by line L2, even after the current convergence period (period t1-t2) has elapsed, the battery current does not converge to within ±10% of the target current value. As shown by lines L1 and L2, the battery current waveform during period t1-t2 differs between normal and abnormal conditions.

[0029] From the measured data shown in Figure 3, the first and second datasets used for the first training can be obtained. The first dataset includes training image A, I1, I2, and ground truth data indicating no slack. Training image A is a representation of the portion of the battery current waveform shown by line L1 within the period t1 to t2, using the aforementioned image format. The second dataset includes training image B, I1, I2, and ground truth data indicating slack. Training image B is a representation of the portion of the battery current waveform shown by line L2 within the period t1 to t2, using the aforementioned image format. In this embodiment, the battery current waveform during the current convergence period is further measured in a manner similar to the example shown in Figure 3, thereby obtaining the necessary number of datasets for both normal and abnormal conditions, and the neural network is trained using the obtained datasets. This generates a first trained model with high judgment accuracy.

[0030] In detail, the ECU500 controls the battery current to follow the fluctuating current target value. When the current target value decreases, the battery current changes to fall below the current target value (undershoot) before rising and approaching the current target value. This makes the battery current value prone to fluctuations. Methods that detect loosening of fasteners based on whether or not the battery current value exceeds a threshold are prone to misjudgments due to fluctuations in the battery current value during normal operation. In contrast, the first trained model determines the presence or absence of loosening of fasteners based on the shape drawn by the battery current value (battery current waveform). This suppresses misjudgments due to fluctuations in the battery current value during normal operation. Also, as shown in Figure 3, the battery current waveform during the current convergence period differs between normal and abnormal operation. Therefore, the first trained model, which has been machine-learned as described above, can determine the presence or absence of loosening of fasteners early and with high accuracy.

[0031] Figure 4 is a time chart illustrating the training images used in the second learning stage. In Figure 4, line L3 shows data under normal conditions. Lines L4 to L6 each show data under abnormal conditions. Lines L3 to L6 each show the progression of the battery current (current of cell 10). Line L20 shows the progression of the current target value.

[0032] In the example shown in Figure 4, the period from t3 to t4 corresponds to the target value stabilization period. During the target value stabilization period, the current target value remains constant. If the current target value is changed, the target value stabilization period begins after the time required for the battery current to converge from the time of the change in the current target value has elapsed. For example, in the time chart shown in Figure 3, the period from t2 onwards corresponds to the target value stabilization period. A portion of the target value stabilization period is extracted and set as the period t3-t4.

[0033] As shown by line L20, during period t3 to t4, the current target value is maintained at the third target value (hereinafter referred to as "I3"). As shown by line L3, under normal conditions, the battery current during the target value stabilization period (period t3 to t4) hardly fluctuates and follows the current target value. In contrast, under abnormal conditions, as shown by lines L4 to L6, the battery current fluctuates violently. This fluctuation is thought to be due to the movement of the conductor member 13 (busbar) when the fastening loosens. The battery current waveform during period t3 to t4 is different between normal and abnormal conditions. The length of period t3 to t4 may be the same as the length of period t1 to t2 shown in Figure 3.

[0034] The third to sixth datasets used in the second learning process can be obtained from the measured data shown in Figure 4. The third dataset includes training image C, I3, and ground truth data showing no slack. Training image C represents the battery current waveform during the target value stabilization period indicated by line L3 in the aforementioned image format. The fourth dataset includes training image D, I3, and ground truth data showing slack occurrence. Training image D represents the battery current waveform during the target value stabilization period indicated by line L4 in the aforementioned image format. The fifth dataset includes training image E, I3, and ground truth data showing slack occurrence. Training image E represents the battery current waveform during the target value stabilization period indicated by line L5 in the aforementioned image format. The sixth dataset includes training image F, I3, and ground truth data showing slack occurrence. Training image F represents the battery current waveform during the target value stabilization period indicated by line L6 in the aforementioned image format. In this embodiment, the battery current waveform during the target value stabilization period is further measured in a manner similar to the example shown in Figure 4, thereby obtaining the necessary number of data sets for both normal and abnormal conditions, and the neural network is subjected to a second training using the obtained data sets. This generates a second trained model with high judgment accuracy.

[0035] More specifically, depending on the condition of vehicle 1000 (including road surface conditions), the battery current may momentarily show abnormal values ​​even under normal conditions, as shown by the dashed line L3A in Figure 4. Methods that detect loosening of fasteners based on whether the battery current value exceeds a threshold are prone to misjudgments due to momentary abnormal values ​​under normal conditions. In contrast, the second trained model determines the presence or absence of loosening of fasteners based on the shape of the battery current value (battery current waveform). This suppresses misjudgments caused by momentary abnormal values ​​under normal conditions. Furthermore, as shown in Figure 4, the battery current waveform during the target value stabilization period differs between normal and abnormal conditions. Therefore, the second trained model, which has been machine-learned as described above, can determine the presence or absence of loosening of fasteners early and with high accuracy.

[0036] As described above, the first and second trained models are generated by the first and second training processes, respectively. Supervised machine learning (for example, shape learning related to battery current waveforms) adjusts the weights W1 between the input layer Nx and the hidden layer Ny, and the weights W2 between the hidden layer Ny and the output layer Nz, so that the target output of the neural network matches the actual output. By repeatedly adjusting the weights W1 and W2 using the training signal, the judgment accuracy of the neural network can be improved. The generated first and second trained models are stored in the memory device 520, as shown in Figure 2.

[0037] The first trained model is configured to output information indicating whether or not loosening has occurred in at least one fastening point in the battery stack (hereinafter referred to as "fastening state information") when it receives the battery current waveform during the current convergence period expressed in the aforementioned image format (hereinafter referred to as "first current waveform image"), the target current value before the decrease, and the target current value after the decrease as input. The second trained model is configured to output fastening state information when it receives the battery current waveform during the target value stabilization period expressed in the aforementioned image format (hereinafter referred to as "second current waveform image") and the target current value during the target value stabilization period as input. For example, each of the first and second trained models outputs either fastening state information "0, 1" indicating no loosening or fastening state information "1, 0" indicating loosening to the output layer Nz, depending on the determination result. In addition, each of the first and second trained models may output "0, 0" or "1, 1" to the output layer Nz if determination is not possible (if the presence or absence of loosening cannot be determined with sufficient accuracy). This configuration makes it easier to improve the judgment accuracy of each trained model. However, it is not limited to this configuration; in the form where the output layer Nz has one node, the neural network may be trained (first training, second training) so that the output layer Nz outputs "0" when there is no slack and outputs "1" when slack occurs.

[0038] The vehicle 1000 shown in Figure 1 is configured to run in one of several driving modes selected from a range of driving modes. For example, the vehicle 1000 is configured to run in a first driving mode in which it is driven only by the MG22 (motor), a second driving mode in which it is driven by both the MG22 and the engine 23, and a third driving mode in which it is driven only by the engine 23. The ECU 500 may switch between the first to third driving modes depending on the status of the vehicle 1000.

[0039] Furthermore, the vehicle 1000 is configured to perform external charging of the battery pack 100 (charging using power supplied from outside the vehicle) while parked. The inlet 410 is configured to be connectable to a charging cable of a power supply facility located outside the vehicle. The charger 420 performs AC / DC conversion. During external charging, AC power is input to the charger 420 from outside the vehicle via the inlet 410, and the ECU 500 controls the charger 420. The charger 420 converts the AC power to DC power according to control commands from the ECU 500 and outputs the DC power to the battery pack 100. This charges each battery contained in the battery pack 100.

[0040] Vehicle 1000 further includes various sensors (position sensor, outside temperature sensor, vehicle speed sensor, odometer, engine status sensor, monitoring unit 100a, etc.) that detect the status of vehicle 1000 in real time. The engine status sensor includes various sensors that detect the status of engine 23 (e.g., fresh air volume, intake pressure, exhaust pressure, engine rotational speed, and engine coolant temperature) in real time. The monitoring unit 100a is located, for example, in the battery pack 100. The monitoring unit 100a includes various sensors that detect the status of each cell 10 (e.g., voltage, current, and temperature). The monitoring unit 100a and ECU 500 may function as a BMS (Battery Management System). In this embodiment, all cells 10 included in the battery pack 100 are connected in series, and the same amount of current flows through all cells 10. Therefore, one current sensor may be shared by all cells 10. However, it is not limited to this, and the battery pack 100 may include multiple cells connected in parallel. A current sensor may be provided for each cell 10.

[0041] The ECU 500 sequentially determines a target current value based on at least one of the following: a request related to the operation of the vehicle 1000 (for example, a request for acceleration, deceleration, or steering from the user or an automated driving system), a request related to the external charging of the battery pack 100 (for example, a request from a power supply facility outside the vehicle), and the state of the vehicle 1000 detected by the various sensors described above. The ECU 500 then performs charge and discharge control of each cell 10 so that the battery current approaches the determined target current value.

[0042] Figure 5 is a flowchart showing the process related to fastening loosening detection by the ECU500. The process flow F1 shown in Figure 5 is repeatedly executed by the ECU500. In the flowchart, "S" represents a step.

[0043] In processing flow F1, the ECU 500 determines in S11 whether the current target value has decreased. The ECU 500 may determine that the current target value has decreased if a predetermined control for decreasing the current target value has been executed. In this embodiment, if the vehicle switches from the first driving mode to the second driving mode, the determination in S11 is YES. Also, if a charge limit is implemented to protect the cell 10 during external charging, the determination in S11 is YES. Furthermore, the ECU 500 may determine whether the current target value has decreased based on whether the value obtained by subtracting the current value of the current target value from the previous value of the current target value exceeds a predetermined value. In this embodiment, even if the previous value of the current target value (current target value before decrease) is greater than the current value of the current target value (current target value after decrease), if the difference between the two (degree of decrease) is small, the determination in S11 is NO.

[0044] If it is determined that the current target value has decreased (YES in S11), the ECU 500 performs a fastening loosening determination in S13 using the first trained model. Specifically, the ECU 500 acquires a first current waveform image. The battery current waveform during the current convergence period is detected by the current sensor included in the monitoring unit 100a. The processor 510 of the ECU 500 then inputs the obtained first current waveform image, the current target value before the decrease, and the current target value after the decrease into the first trained model. As a result, fastening status information is output from the first trained model. After that, the process proceeds to S15.

[0045] In S15, the ECU 500 determines whether the obtained fastening status information indicates that loosening has occurred. If the fastening status information indicates that loosening has occurred (YES in S15), the ECU 500 notifies the user in S16 that an abnormality (specifically, fastening loosening) has occurred in the battery pack 100. The ECU 500 may also notify the user terminal (e.g., an in-vehicle terminal or a mobile terminal) of the abnormality in the battery pack 100. The user terminal that receives this notification may display a message to the user indicating the abnormality in the battery pack 100. In the subsequent S17, the ECU 500 performs a predetermined fail-safe process. The ECU 500 stores information indicating that fastening loosening has occurred in the battery pack 100 in the storage device 520. The ECU 500 also shuts off the input / output power of the battery pack 100 by, for example, setting a relay (not shown) provided in the input / output section of the battery pack 100 to an open state. If vehicle 1000 is running in the first or second driving mode, ECU 500 switches to the third driving mode and continues to drive vehicle 1000 using engine 23. If vehicle 1000 is running in the third driving mode, ECU 500 continues to drive vehicle 1000 in the third driving mode. If the battery pack 100 is being externally charged, ECU 500 notifies the power supply equipment to stop external charging and stops external charging of the battery pack 100. Then, ECU 500 prohibits future use of the first driving mode, the second driving mode, and external charging. Once processing S17 is executed, processing flow F1 ends.

[0046] On the other hand, if the fastening status information indicates no loosening or inability to determine (NO in S15), the process returns to the first step (S11). Also, if it is determined that the current target value has not decreased (NO in S11), the ECU 500 determines in S12 whether or not the target value stabilization period has begun. The ECU 500 may determine whether or not the target value stabilization period has begun based on whether or not the current convergence period has elapsed since the most recent change in the current target value. In this embodiment, if the current convergence period related to the decrease in the current target value has elapsed since the determination of YES in S11, the determination of YES in S12 is made. Also, the ECU 500 may determine that the target value stabilization period has begun if the current target value has not changed for a predetermined period of time.

[0047] If it is determined that the target value stabilization period has not yet begun (NO in S12), the process returns to S11. On the other hand, if it is determined that the target value stabilization period has begun (YES in S12), the ECU 500 performs a fastening loosening determination in S14 using the second trained model. Specifically, the ECU 500 acquires a second current waveform image. The battery current waveform during the target value stabilization period is detected by the current sensor included in the monitoring unit 100a. The processor 510 of the ECU 500 then inputs the obtained second current waveform image and the current target value (constant value) during the target value stabilization period into the second trained model. As a result, fastening status information is output from the second trained model. After that, the process proceeds to S15. In S15, the presence or absence of an abnormality (fastening loosening) is determined based on the fastening status information output from the second trained model. If an abnormality is determined, the processes in S16 and S17 are executed.

[0048] As described above, the ECU 500 (fastening anomaly determination device) according to this embodiment includes a storage device 520 that stores a first trained model and a second trained model. Each of the first and second trained models is machine-trained to output output information indicating whether or not fastening loosening has occurred in the fastening structure when predetermined input information is input. The input information includes waveform data (first current waveform image, second current waveform image) showing the waveform of the current flowing through the fastening structure. The ECU 500 is configured to acquire waveform data using a current sensor (monitoring unit 100a), input the input information including the acquired waveform data into a trained model (first trained model or second trained model), and determine whether or not fastening loosening has occurred in the fastening structure (fastening part of cell 10) based on the output information output from the trained model that has received the input information. With such a fastening anomaly determination device, it is possible to determine with high accuracy whether or not fastening loosening has occurred in the fastening structure using the trained models.

[0049] In the above embodiment, a first trained model is obtained by performing first training on an untrained neural network, and a second trained model is obtained by performing second training on another untrained neural network. In both the first and second trained models, the output information is fastening status information. On the other hand, the input information differs between the first and second trained models. The input information for the first trained model includes a first current waveform image, the current target value before the decrease, and the current target value after the decrease. The input information for the second trained model includes a second current waveform image and the current target value during the target value stabilization period. The input information and output information correspond to explanatory variables and objective variables, respectively. When the current target value decreases, the ECU500 uses the first trained model to determine whether or not fastening loosening has occurred in the fastening structure (S13 in Figure 5), and when the current target value remains constant, it uses the second trained model to determine whether or not fastening loosening has occurred in the fastening structure (S14 in Figure 5). With this configuration, training can be performed efficiently using situation-specific training data. However, this is not the only way; a trained model possessing the functionality of both the first and second trained models may be generated by performing both the first and second training processes on a single untrained neural network.

[0050] In the modified version, supervised machine learning is performed on a neural network using a training image showing the waveform of the battery current value in a predetermined interval, a training image showing the waveform of the target current value in a predetermined interval, the moving average of the battery current, the moving average of the target current value, the standard deviation of the battery current, the standard deviation of the target current value, and the ground truth data. The trained model generated by this training (hereinafter referred to as the "trained model according to the modified version") is stored in the storage device 520 of the ECU 500 in place of the first and second trained models described above. The ECU 500 executes the processing flow F2 shown in Figure 6 instead of the processing flow F1 shown in Figure 5.

[0051] Figure 6 is a flowchart showing a modified version of the processing flow shown in Figure 5. Processing flow F2 is the same as processing flow F1 shown in Figure 5, except that S13A is used instead of S11-S14 (Figure 5).

[0052] In S13A, the ECU 500 performs fastening loosening determination using a trained model related to the modified example. Specifically, the ECU 500 acquires the waveform of the battery current value, the waveform of the current target value, the moving average of the battery current, the moving average of the current target value, the standard deviation of the battery current, and the standard deviation of the current target value for a predetermined interval (for example, the most recent 5-second interval). The ECU 500 then converts the waveform of the battery current value and the waveform of the current target value into images in a predetermined format. The processor 510 of the ECU 500 then inputs the waveform (image) of the battery current value, the waveform (image) of the current target value, the moving average of the battery current, the moving average of the current target value, the standard deviation of the battery current, and the standard deviation of the current target value for the predetermined interval into the trained model related to the modified example. As a result, fastening status information is output from the trained model related to the modified example. After that, the process proceeds to S15. According to the processing flow F2, fastening loosening determination is repeatedly performed regardless of the behavior of the current target value. That is, fastening loosening determination is performed even when the current target value increases. Furthermore, using the trained model related to the above modification, it is possible to determine with high accuracy whether or not loosening has occurred in the fastening structure (the fastening portion of cell 10).

[0053] In the trained model related to the above modification, both moving average and standard deviation are used as features (input information) to improve the accuracy of the judgment. However, these are not essential inputs, and at least one of the moving average and standard deviation may be omitted.

[0054] The neural network training may be performed sequentially, for example, by a server on the cloud. The server may then update the trained model in the ECU500 via OTA (Over The Air). The training method is not limited to supervised machine learning; it may also be unsupervised machine learning.

[0055] The fastening structure whose presence or absence of loosening is determined by the fastening abnormality determination device (ECU500) described above is not limited to the fastening parts of a battery (negative terminal 11, positive terminal 12, and conductor member 13), but can be any structure in which multiple fastened parts are fastened together with fastening components.

[0056] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of the present invention is indicated by the claims rather than by the description of the embodiments above, and all modifications within the meaning and scope equivalent to the claims are intended to be included. [Explanation of symbols]

[0057] 10 Cell, 11 Negative terminal, 12 Positive terminal, 13 Conductor component, 100 Battery pack, 500 ECU, 520 Memory device, 1000 Vehicle, E4 Nut, E5 Bolt.

Claims

1. A fastening abnormality determination device for determining whether or not a fastening loosening has occurred in a fastening structure in which multiple fastened parts are fastened together with fastening parts, The fastening abnormality determination device includes a storage device that stores a trained model that has been trained to output output information indicating whether or not a fastening loosen has occurred in the fastening structure when predetermined input information is received. The input information includes waveform data showing the waveform of the current flowing through the fastening structure. The fastening abnormality determination device is configured to acquire waveform data using a current sensor, input the acquired waveform data into the trained model, and determine whether or not a fastening loosening has occurred in the fastening structure based on the output information output from the trained model.

2. The fastening abnormality detection device is configured to control the current flowing through the fastening structure so that the current flowing through the fastening structure approaches a target value. The fastening abnormality determination device according to claim 1, wherein the input information further includes the target value.

3. The memory device stores the first trained model and the second trained model as the trained models. The fastening abnormality determination device is, When the target value decreases, the first trained model is used to determine whether or not the fastening structure has become loose. The fastening abnormality determination device according to claim 2, configured to determine whether or not fastening loosening has occurred in the fastening structure using the second trained model when the target value is constant.

4. A vehicle comprising a fastening abnormality determination device according to any one of claims 1 to 3 and the fastening structure, The aforementioned vehicle is A motor that drives the vehicle using power output from a battery installed in the vehicle, An internal combustion engine that uses the combustion energy of fuel to drive the vehicle, Equipped with, The fastening structure is a fastening part for the battery, in a vehicle.

5. The vehicle according to claim 4, wherein, while the vehicle is being driven by the motor, if the fastening abnormality determination device determines that a fastening loosening has occurred in the fastening structure, the fastening abnormality determination device is configured to switch from driving by the motor to driving by the internal combustion engine and continue driving the vehicle with the internal combustion engine.

Images