Electronic device
A deep learning-based monitoring system for lighting devices detects signs of failure in power supply circuits by analyzing waveforms, improving accuracy and enabling proactive maintenance to prevent unexpected malfunctions.
Patent Information
- Application Number
- JP2020165574
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2020-09-30
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2040-09-30
AI Technical Summary
Existing methods for detecting failures in lighting devices, such as those in road tunnels, are labor-intensive and prone to false determinations due to transient noise or reliance on cumulative lighting time, leading to potential safety risks and high costs from unexpected malfunctions.
A monitoring system that analyzes the waveform of the power supply circuit using deep learning to detect signs of failure before they occur, reducing false determinations by comparing monitored waveforms against a learned model and transmitting abnormality information to a host device.
Accurately detects signs of failure with reduced false alarms, allowing for proactive maintenance and minimizing the costs and disruptions associated with sudden equipment failures.
Smart Images

Figure 0007577498000001 
Figure 0007577498000002 
Figure 0007577498000003
Abstract
Description
[Technical field]
[0001] The present invention relates to an electronic device equipped with a power supply circuit, such as a lighting device. [Background technology]
[0002] For example, malfunctions in lighting devices installed in road tunnels and other electronic devices are only discovered when the lighting stops working, making it difficult to grasp such malfunctions in advance. If replacement work is performed after the malfunction is discovered, unexpected costs will be incurred due to traffic restrictions for the work, and the procedures for raising the funds will be time-consuming, making it difficult to respond quickly.
[0003] Currently, in order to grasp the status of lighting devices in advance, regular patrols are required, which is very labor-intensive, and as a result, lighting devices are sometimes left in an unlit state, compromising safety.
[0004] As a method for detecting a malfunction of a lighting device, Patent Document 1 describes that a malfunction detection unit generates a malfunction detection value when the input power, input current, output power, or output current detected by a sensor exceeds a predetermined range. Patent Document 1 also describes that a malfunction detection value may be generated when the cumulative lighting time exceeds a predetermined time. [Prior art documents] [Patent documents]
[0005] [Patent Document 1] JP 2019-186114 A Summary of the Invention [Problem to be solved by the invention]
[0006] In the case of the invention described in Patent Document 1, the failure detection unit estimates a failure based on a predetermined range or cumulative lighting time. In other words, in Patent Document 1, it estimates whether or not a failure has already occurred, and it is not possible to take prompt action as described above after a failure has occurred. Furthermore, when threshold determination is made based on upper and lower limit values such as a predetermined range, there is a risk of erroneous determination due to transient noise, etc. Furthermore, the cumulative lighting time does not directly determine the presence or absence of a malfunction such as a failure.
[0007] SUMMARY OF THE PRESENT EMBODIMENTS An object of the present invention is to provide an electronic device that can reduce erroneous judgments and accurately detect abnormal conditions such as signs of a breakdown. [Means for solving the problem]
[0008] The invention described in claim 1, made to solve the above problem, is an electronic device comprising a monitoring unit that monitors the waveform of an object to be monitored in a power supply circuit, and a communication unit that transmits the monitoring result by the monitoring unit to the outside, and when the monitoring unit detects an abnormality in the power supply circuit based on the waveform of the object to be monitored, the monitoring unit causes the communication unit to transmit information indicating the detection of an abnormality. Effect of the Invention
[0009] According to the present invention, when the monitoring unit detects an abnormality in the waveform of the object to be monitored, it causes the communication unit to transmit information indicating the abnormality detection, thereby reducing erroneous judgments and making it possible to detect abnormal conditions such as signs of failure with greater accuracy than if an upper limit value or the like were simply set. [Brief description of the drawings]
[0010] [Figure 1] FIG. 2 is a functional configuration diagram of a lighting device according to an embodiment of the present invention. [Diagram 2] 2 is a graph showing the relationship between the output voltage value and operation time of the constant current control unit shown in FIG. 1. [Diagram 3] 4 is an example of an output voltage waveform in a normal state. [Figure 4] 4 is an example of an output voltage waveform in an abnormal state. [Diagram 5] These are examples of items that need to be monitored during learning and inference. [Figure 6] FIG. 13 is an explanatory diagram for forming a two-dimensional array of acquired measurement data. [Figure 7] FIG. 7 is an explanatory diagram of imaging of the two-dimensional array in FIG. 6. [Figure 8] FIG. 1 is an explanatory diagram of learning in deep learning. [Figure 9] FIG. 1 is an explanatory diagram of inference in deep learning. [Figure 10] 4 is a flowchart of an abnormality detection operation in the current monitoring unit shown in FIG. BEST MODE FOR CARRYING OUT THEINVENTION
[0011] Hereinafter, an embodiment of the present invention will be described with reference to Fig. 1 to Fig. 10. Fig. 1 is a functional configuration diagram of a lighting device as an electronic device according to an embodiment of the present invention.
[0012] The lighting device 1 includes a communication unit 2, a constant current control unit 3, an output monitoring unit 4, a current monitoring unit 5, and an LED light source 6. The lighting device 1 is installed, for example, as lighting in a road tunnel (tunnel lighting).
[0013] In this embodiment, the lighting device is described as a road tunnel lighting device, but the lighting device may be other road lighting devices such as street lights, or lighting devices other than road lighting devices. In other words, the lighting device may be a lighting device that has a built-in communication function and communicates with a higher-level device that can control the lighting device itself.
[0014] The communication unit 2 is connected to a higher-level device such as a dimmer via a communication line. The communication unit 2 receives dimming information (information on the luminance of the light source) of the LED light source 6 from the higher-level device, and transmits monitoring results (abnormality information) of the current monitoring unit 5, which will be described later, to the higher-level device. The communication unit 2 is provided with identification information such as an ID in advance, so that it can be distinguished from other lighting devices 1 in communication with the higher-level device. The communication line connected to the higher-level device can be any type, such as power line communication or wireless communication.
[0015] The constant current control unit 3 is composed of circuits such as a rectifier circuit and a constant current drive circuit for controlling the LED light source 6 at a constant current. It is known that the light emitting diodes constituting the LED light source 6 are generally driven at a constant current. The constant current control unit 3 receives AC power (AC input) from a power receiving facility or the like installed outside the road tunnel, converts the AC power to DC, and generates and outputs a current (output current) for driving the light emitting diodes of the LED light source 6 at a constant current. In other words, the constant current control unit 3 functions as a power supply circuit that supplies a current for the LED light source 6 to emit light.
[0016] The output monitoring unit 4 is a sensing circuit added to the output circuit of the constant current control unit 3. This sensing circuit is generally composed of a resistive element. The monitoring result of the output monitoring unit 4 is output to the current monitoring unit 5.
[0017] The current monitoring unit 5 controls the constant current control unit 3 so that the current value is based on the dimming information received by the communication unit 2. The current monitoring unit 5 also feedback controls the operation of the constant current control unit 3 based on the detection result of the output monitoring unit 4. This makes it possible to appropriately dim the brightness of the LED light source 6 by setting the output current value of the constant current control unit 3 to a current value according to the dimming degree.
[0018] Furthermore, the current monitoring unit 5 monitors the output current waveform, etc. of the constant current control unit 3, and if the waveform, etc. is abnormal, outputs abnormality information to the communication unit 2. In this embodiment, abnormality does not refer to a state that has led to a breakdown, but rather to a state that does not lead to a breakdown but shows signs of leading to a breakdown. In other words, the current monitoring unit 5 functions as a monitoring unit that monitors the output current waveform, etc. (waveform to be monitored) in the constant current control unit 3 (power supply circuit), and if an abnormality in the constant current control unit 3 (power supply circuit) is detected based on the output current waveform, etc. (waveform to be monitored), it causes the communication unit 2 to transmit abnormality information (information indicating abnormality detection).
[0019] As described above, the LED light source 6 uses a light emitting diode as a light source. The LED light source 6 is driven (to emit light) by a drive current (output current) supplied from the constant current control unit 3.
[0020] Here, the detection of an abnormality in the current waveform in this embodiment will be described. Fig. 2 is a graph showing the relationship between the output voltage value of the constant current control unit 3 and the operation time. As shown in Fig. 2, it is known that the output voltage value of the constant current control unit 3 gradually decreases due to the deterioration of the circuit components constituting the constant current control unit 3.
[0021] In Fig. 2, when the voltage drops below Vth, the lighting device 1 stops functioning (failure state). Conventionally, a failure state was detected by detecting when the voltage dropped below Vth, but in the case of road lighting such as road tunnel lighting, if replacement is done after a failure occurs, unexpected costs arise due to traffic restrictions for work, and the procedures for finding the funds are time-consuming, making it difficult to respond quickly.
[0022] Therefore, in this embodiment, a symptom of a failure is detected before it leads to a failure and notified in advance. For example, assume that the output voltage is normal at t1 in Fig. 2, and the output voltage approaches Vth at t2, a time that has passed since t1. At t2, the output voltage is still operating normally, but has deteriorated more than at t1 and is approaching the time of failure. Therefore, if a symptom of a failure can be detected at t2, that is, before it reaches Vth, it becomes possible to take action such as arranging replacement or other work before the failure occurs.
[0023] The symptom of a fault is detected based on the waveform of the output current from the constant current control unit 3 (output current waveform). The following description will be given with reference to Figs. 3 and 4. Fig. 3 shows an example of an output current waveform in an initial state (normal state). As shown in Fig. 3, in a normal state, the amplitude, frequency, etc. are substantially constant. On the other hand, in an abnormal state, as shown in Fig. 4, the amplitude, frequency, etc. change suddenly and the waveform becomes distorted. Such a phenomenon may be caused, for example, by an increase in ripple due to the deterioration of the electrolytic capacitor constituting the constant current control unit 3. It may also be caused by the failure of a component other than the electrolytic capacitor. That is, an abnormality caused by the deterioration of a component constituting the constant current control unit 3 (power supply circuit) is detected based on the output current waveform, etc. (waveform to be monitored).
[0024] Therefore, in this embodiment, an abnormal state is detected (inferred) based on a learning model generated by learning the waveform of the monitored object under normal conditions. The method is described below. First, as shown in FIG. 5, data is acquired for the monitored items. In FIG. 5, as examples, each item in the constant current control unit 3, such as output current, output voltage, PFC (power factor correction circuit) voltage, auxiliary winding voltage, microcomputer power supply voltage, switching waveform, transformer temperature, FET temperature, etc. Of course, all of these items (waveforms) are not essential, and other items may be included. However, in order to make a two-dimensional array as described later, multiple items are necessary. Note that, for comparison, normal and abnormal waveforms are shown in FIG. 5, but data may be collected during learning only under normal conditions.
[0025] For data collected as shown in Figure 5, the waveforms differ depending on the sampling start timing, so it is necessary to delete the time information. Therefore, as shown in Figure 6, the above items are expressed in a two-dimensional array in chronological order. In the array in Figure 6, the vertical direction represents items a to h in Figure 5, and the horizontal direction represents the time series. In Figure 6, the waveforms, etc. of each item are collected for n consecutive sampling times from (0) to (n-1).
[0026] Next, FFT (Fast Fourier Transform) is performed to remove the time information and leave the amplitude information, frequency information, etc. The result of FFT is shown in Figure 7. In Figure 7, the result of FFT on a(0) in Figure 6 is shown as A(0).
[0027] Then, the two-dimensional array generated as shown in Fig. 7 is treated as image data and is trained by deep learning. Learning by deep learning will be described with reference to Fig. 8. In Fig. 8, input data (image data) is input to a neural network 50. The neural network 50 has cascaded layers 51 and weight parameters 52 that each layer 51 has.
[0028] Layer 51 is a filter for extracting features from input data, and generally a two-dimensional convolutional layer or a fully connected layer is used. Layer 51 has weight parameters 52, and in the case of a two-dimensional convolutional layer or a fully connected layer, a convolutional operation is performed with the input data. In layer 51, a convolutional operation is performed with the input data and the weight parameters 52 of layer 51. The result of this operation becomes the input of the next layer 51, and a convolutional operation is performed with the weight parameters 52 of the next layer 51. The same process is repeated for the number of layers 51, and the output of the last layer 51 is converted into a probability to obtain a predicted value 53.
[0029] A predicted value 53 is output from the neural network 50. A loss function 55 calculates how far this predicted value 53 is from a true value 54. An error 56 is obtained as the output of the loss function 55. An optimizer 57 then determines how to proceed with learning based on the error 56. Generally, an algorithm called backpropagation is implemented to determine how to adjust the weight parameters 52 based on the error, and the weight parameters 52 are updated.
[0030] A learning model is generated by repeating such learning and adjusting the weighting parameters 52 so as to reduce the error 56. This learning operation is performed by a computer or the like (not shown). That is, the learning model is generated by learning from images of measurement results sampled for a predetermined period of time for multiple monitored objects, which are arranged in a two-dimensional array.
[0031] The generated learning model is stored in the current monitoring unit 5. As shown in FIG. 9, the part stored in the current monitoring unit 5 is the part related to the neural network 50. That is, the part related to the neural network 50 in FIG. 9 becomes the learning model. Then, when measurement data (image data) is input during inference, a predicted value 53 is output. In this embodiment, the coincidence between the data in a normal state and the measurement data is output as a probability as the predicted value 53. For example, it can be determined that a coincidence of 0% is a failure, 30% is likely to break, 70% is deteriorated, and 100% is normal. That is, in this embodiment, a sign of failure can be captured in a state of 30% or 70%. Therefore, it is possible to inform the timing of replacing the power supply. In addition, since the correlation of each data can be obtained by forming a two-dimensional array, the accuracy of determining an abnormal state such as a failure is improved.
[0032] In the above explanation, normal measurement data is learned, but abnormal (fault) measurement data may be learned. In this case, the above probability of 0% is normal and 100% is fault. Furthermore, abnormal measurement data is not limited to faults and may be a waveform of a target degradation state.
[0033] In this embodiment, as a method other than learning, the deterioration characteristics may be stored in advance and compared with the actual output current waveform, etc. to detect an abnormal state. The deterioration characteristics to be stored are the characteristics before a failure occurred (for example, the state at time t2 in FIG. 2). Examples of the deterioration characteristics include the feature quantities of the amplitude and frequency characteristics of the waveform at the time of deterioration, or the waveform itself. In this method, it is sufficient that there is one or more waveforms (items) to be monitored. In the case of multiple items, if any of the waveforms to be monitored is determined to be abnormal, it may be output to a higher-level device.
[0034] Furthermore, the current monitoring unit 5 may include information on the approximate remaining time until the malfunction occurs in the abnormality information. The remaining time can be calculated, for example, by measuring the actual device until it breaks down, and acquiring the characteristics of the output waveform during that time (such as the state of change). The acquired characteristics are then set in the current monitoring unit 5. The set characteristics are then compared with the characteristics of the current output current waveform, and the remaining time can be predicted based on whether they match or are similar. For example, the characteristics of the output waveform can be stored every 100 hours until it breaks down, and the remaining time can be calculated from the difference between the time corresponding to the closest waveform and the time of the breakdown.
[0035] In addition, it is preferable to detect the above-mentioned abnormality based on multiple periods (several consecutive waveforms) of the waveform per sample rather than one period. This makes it possible to detect abnormalities whose features are difficult to extract based on one period.
[0036] Furthermore, since the deterioration of electronic components progresses over time, it may be determined that the possibility of the above-mentioned abnormality occurring is low when the lighting device 1 is installed, and the current monitoring unit 5 may perform abnormality determination after a predetermined time has passed since the installation of the lighting device 1. The time to start abnormality determination may be appropriately determined, such as about half the expected lifespan. For example, if the expected lifespan is 40,000 hours, abnormality determination may be started 20,000 hours after installation. The start time of abnormality determination may be detected by measuring time using an internal timer, or may be determined by receiving information indicating the start of abnormality determination from a higher-level device via communication.
[0037] Furthermore, the abnormality determination may be performed only when an instruction is received from a higher-level device. By performing the determination based on an instruction from a higher-level device, the current monitoring unit 5 does not need to constantly detect abnormalities, and the processing load can be reduced because the current monitoring unit 5 does not need to detect the abnormality by itself.
[0038] In the above explanation, the abnormality is mainly a symptom of failure due to aging of electronic components such as electrolytic capacitors, but for example, an abnormality due to intermittent noise may be detected in a similar manner and abnormality information may be transmitted to a higher-level device. In this case, it is preferable not to operate the current monitoring unit 5 to determine an abnormality after a predetermined time has elapsed since the installation of the lighting device 1.
[0039] The above-mentioned anomaly detection operation in the current monitoring unit 5 will be described with reference to the flowchart in Fig. 10. First, it is determined whether to start anomaly detection (step S1). In step S1, as described above, it is determined whether a trigger for performing an anomaly detection operation has been issued by an internal timer or an instruction from a higher-level device. If a trigger has not been issued (step S1; No), step S1 is repeated. Note that step S1 can be omitted if the operation of determining anomaly by the current monitoring unit 5 is not to function until a predetermined time has elapsed since the installation of the lighting device 1.
[0040] On the other hand, if a trigger has been issued (step S1; Yes), measurement data such as the output current waveform is acquired from the constant current control unit 3 (step S2). This measurement data needs to have the same number of samples (the same number of n) as during learning. Then, it is determined whether an abnormality has been detected in the acquired measurement data (step S3). If no abnormality is detected (step S3; No), the process returns to step S2 and measurement data is acquired again.
[0041] On the other hand, if an abnormality is detected (step S3; Yes), abnormality information is output to the communication unit 2 (step S4). This abnormality information includes its own ID, information that an abnormality has been detected (such as the above-mentioned probability), the date and time of detection, etc. It may also include information regarding the approximate remaining time until the above-mentioned function shutdown occurs.
[0042] The abnormality information is not limited to the information indicating the above-mentioned failure symptoms, but may also include information indicating a failure state. For example, if a failure is detected after the information indicating the failure symptoms is transmitted, the information indicating the failure state is also transmitted. The failure state is not limited to the above-mentioned probability of 0%, and may be determined by a known conventional method.
[0043] According to this embodiment, the lighting device 1 includes an LED light source 6, a constant current control unit 3 that supplies a drive current to the LED light source 6, a current monitoring unit 5 that monitors an output current waveform in the constant current control unit 3, and a communication unit 2 that transmits abnormality information indicating that an abnormality has been detected by the current monitoring unit 5 to an external higher-level device. When the current monitoring unit 5 detects an abnormality in the output current waveform, the communication unit 2 transmits information indicating the abnormality information. In this manner, when the current monitoring unit 5 detects an abnormality in the output current waveform, the communication unit 2 transmits the abnormality information. This makes a judgment based on the waveform rather than simply a value set with an upper limit value, etc., and thus makes it possible to accurately detect an abnormal state such as a symptom of a failure with fewer erroneous judgments.
[0044] In addition, the monitoring unit detects signs of failure due to deterioration of the components that make up the constant current control unit 3 based on the output current waveform, etc., making it possible to capture signs of failure due to aging and take action before the unit stops functioning.
[0045] Furthermore, by including the output current waveform of the constant current control unit 3 in the waveforms to be monitored, abnormalities occurring when the LED light source 6 is driven can be easily detected.
[0046] In addition, since the current monitoring unit 5 detects abnormalities based on multiple consecutive waveforms, it becomes possible to detect abnormalities based on waveform characteristics such as abnormalities in frequency characteristics, and this allows for more accurate detection of abnormalities than detecting abnormalities based on values at specific times (instantaneous values).
[0047] Moreover, the current monitoring unit 5 detects the abnormality with a learning model that has learned the waveform of the monitored object under normal conditions, so that the abnormality can be detected with higher accuracy. Moreover, the learning model is generated by learning from a two-dimensional array of imaged measurement results sampled for a predetermined period of time for a plurality of monitored items, so that the two-dimensional array makes it possible to obtain the correlation of each data, thereby improving the accuracy of judging an abnormal state.
[0048] The current monitoring unit 5 may start monitoring after a predetermined time has elapsed since installation. In this way, when detecting aged deterioration, monitoring can be started from the point when deterioration has progressed. Therefore, the processing load on the current monitoring unit 5 can be reduced.
[0049] In addition, the brightness of the LED light source 6 can be adjusted, and the communication unit 2 receives information regarding the brightness of the LED light source 6. In this way, individual dimming is possible in road tunnel lighting where multiple lighting devices 1 are installed. Therefore, it is possible to illuminate with an appropriate brightness depending on the installation position of the lighting device 1.
[0050] In the above embodiment, the electronic device is a lighting device, but the present invention is not limited to this. For example, the present invention can be applied to any electronic device that has a power supply circuit and a means for communicating with a higher-level device, such as a computer terminal, a home appliance, or an in-vehicle device.
[0051] The present invention is not limited to the above-described embodiment. In other words, a person skilled in the art can implement the present invention by modifying it in various ways in accordance with conventional knowledge without departing from the gist of the present invention. As long as the electronic device of the present invention is still provided even after such modification, it is of course included in the scope of the present invention. [Explanation of symbols]
[0052] 1 Lighting equipment (electronic equipment) 2. Communications Department 3 Constant current control section (power supply circuit) 5 Current monitoring section (monitoring section) 6 LED light source 50 Neural Networks (Learning Models)
Claims
1. A monitoring unit that monitors a waveform of a monitoring target in a power supply circuit; a communication unit that transmits a monitoring result by the monitoring unit to an outside, When the monitoring unit detects an abnormality in the power supply circuit based on the waveform of the monitored object, the monitoring unit causes the communication unit to transmit information indicating the detection of the abnormality; The monitoring unit detects the abnormality using a learning model that has learned a waveform of the object to be monitored in a normal state or an abnormal state, The electronic device is characterized in that the learning model is generated by learning from a two-dimensional array of imaged measurement results sampled over a predetermined period of time for multiple monitored objects.
2. The electronic device according to claim 1 , wherein the monitoring unit detects the abnormality based on a plurality of successive waveforms.
3. 3. The electronic device according to claim 1, wherein the monitoring unit detects a symptom of a failure due to deterioration of a part based on the waveform of the monitored object.
4. 4. The electronic device according to claim 1, wherein the monitoring unit monitors the waveform of the monitoring target after a predetermined time has elapsed since installation.
5. A light source and the power supply circuit supplying a current for emitting light from the light source, The monitoring unit monitors a waveform of the monitoring target in the power supply circuit.
5. The electronic device according to claim 1, wherein the electronic device is a liquid crystal display.
6. 6. The electronic device according to claim 5, wherein the waveform to be monitored includes an output current waveform supplied to the light source.
Citation Information
Patent Citations
Waveform evaluation device utilizing neural network and method therefor
JP1995013983A
Degradation diagnostic device for DC power unit
JP2004280549A
Marker lamp and marker lamp system
JP2013206852A
Power supply unit and control method therefor
JP2014166059A
Lighting operational system and cloud server
JP2019186114A
