Warning sign detecting device, warning sign detecting method, and program

Abstract

An objective of the present invention is to improve the accuracy of prioritizing maintenance work to be performed on an air conditioner in which a failure is predicted. The present disclosure provides a warning sign detecting device that detects a warning sign of failure of an air conditioning device 40 that is the subject of maintenance work, wherein the warning sign detecting device: acquires operating data from the air conditioning device 40; detects a warning sign of failure on the basis of the operating data acquired from the air conditioning device 40; acquires the air conditioning capability of the air conditioning device 40; on the basis of the air conditioning capability, determines the priority level of maintenance work for the air conditioning device in which a warning sign of failure has been detected; and outputs the priority level together with the warning sign of failure.

Description

Predictive detection device, predictive detection method, program

[0001] This invention relates to a predictive detection device, a predictive detection method, and a program.

[0002] When equipment such as air conditioning systems (hereinafter referred to as "air conditioners") malfunction, it becomes difficult to air condition the space being air-conditioned. Therefore, there is a known technology that detects signs of malfunction from the operating data of air conditioners and performs maintenance work on the air conditioners before a failure occurs.

[0003] Patent Document 1 discloses a technology for detecting multiple signs of failure in a diagnostic target device and for determining the maintenance priority for each of the detected signs.

[0004] Japanese Patent Publication No. 2022-185392

[0005] However, conventional technologies may have low accuracy in determining the priority of maintenance work based on fault prediction. The purpose of detecting signs of failure is to perform maintenance work before a failure occurs. For example, conventional technologies identify parts that show signs of failure based on operating data and determine the priority of maintenance work based on those parts. However, even if a part is important, a failure may not occur early, and the priority of maintenance work may not be high.

[0006] This disclosure provides a technology to improve the accuracy of prioritizing maintenance work performed on air conditioners for which failures are predicted.

[0007] A first aspect of the present disclosure is a predictive maintenance device for detecting signs of failure in an air conditioning system that is subject to maintenance work, the device comprising a control unit, the control unit acquiring operating data from the air conditioning system, detecting signs of failure based on the operating data acquired from the air conditioning system, acquiring the air conditioning capacity of the air conditioning system, determining the priority of maintenance work for the air conditioning system in which signs of failure have been detected based on the air conditioning capacity, and outputting the priority along with the signs of failure.

[0008] According to a first aspect of this disclosure, the accuracy of prioritizing maintenance work performed on air conditioners for which failures are predicted can be improved.

[0009] A second aspect of this disclosure is a fault detection device according to the first aspect, wherein the control unit outputs a time when a fault is predicted to occur when a fault is detected.

[0010] A third aspect of this disclosure is a fault detection device according to the first aspect, wherein the control unit outputs the probability of a fault occurring when a fault precursor is detected.

[0011] A fourth aspect of the present disclosure is a predictive maintenance device according to the second or third aspect, wherein the control unit outputs a predicted air conditioning capacity by inputting the operating data acquired from the air conditioning device into a model that has learned the correspondence between the operating data acquired from the air conditioning device and the air conditioning capacity, calculates the actual air conditioning capacity using the operating data acquired from the air conditioning device, and determines that the air conditioning capacity of the air conditioning device has decreased based on predetermined criteria if the difference between the predicted air conditioning capacity and the actual air conditioning capacity is greater than a standard value.

[0012] A fifth aspect of this disclosure is a predictive maintenance device according to any of the first to fourth aspects, wherein the predictive maintenance device can communicate via a network with a terminal operated by a worker or a user of the air conditioner, and the control unit outputs to the terminal a message proposing maintenance work for the air conditioner when a sign of failure is detected and the air conditioning capacity of the air conditioner is determined to be reduced based on predetermined criteria.

[0013] A sixth aspect of this disclosure is a predictive maintenance device according to any of the first to fifth aspects, wherein when a predictive maintenance is detected for an air conditioner and the control unit is determined to have a reduced air conditioning capacity based on predetermined criteria, the control unit calculates the air conditioning capacity of the air conditioner after the completion of the maintenance work.

[0014] A seventh aspect of this disclosure is a predictive maintenance device according to the sixth aspect, wherein the control unit outputs information regarding the air conditioning capacity of the air conditioning system when it is determined that the air conditioning capacity before the maintenance work is performed has decreased based on predetermined criteria, and when it is determined that the air conditioning capacity calculated after the maintenance work is completed has not decreased based on predetermined criteria.

[0015] An eighth aspect of the present disclosure is a fault detection device according to any of the first to seventh aspects, wherein the control unit sets the priority of maintenance work for an air conditioner in which fault signs are detected and the air conditioning capacity of the air conditioner is determined to have decreased based on predetermined criteria to be higher than the priority of maintenance work for an air conditioner in which fault signs are detected and the air conditioning capacity of the air conditioner is determined not to have decreased based on predetermined criteria.

[0016] A ninth aspect of the present disclosure is a fault detection device according to any of the first to eighth aspects, wherein the control unit determines the rate of decrease in the air conditioning capacity of the air conditioning device when it is determined that the air conditioning capacity of the air conditioning device is decreasing based on predetermined criteria, and if the rate of decrease in the air conditioning capacity is greater than or equal to a threshold, the priority of maintenance work on the air conditioning device in which a fault indicator has been detected is increased compared to when the rate of decrease in the air conditioning capacity is not greater than or equal to a threshold.

[0017] A tenth aspect of the present disclosure is a predictive detection method for an air conditioning system that is subject to maintenance work, wherein the predictive detection device has a control unit, and the control unit performs the following: a process of acquiring operating data from the air conditioning system; a process of detecting a precursor of failure based on the operating data acquired from the air conditioning system; a process of acquiring the air conditioning capacity of the air conditioning system; a process of determining the priority of maintenance work for the air conditioning system in which a precursor of failure has been detected based on the air conditioning capacity; and a process of outputting the priority along with the precursor of failure.

[0018] According to a tenth aspect of this disclosure, the accuracy of prioritizing maintenance work performed on air conditioners for which failures are predicted can be improved.

[0019] An eleventh aspect of this disclosure is a program for causing a predictive maintenance device for detecting signs of failure in an air conditioning system that is subject to maintenance work to execute the following processes: acquiring operating data from the air conditioning system; detecting signs of failure based on the operating data acquired from the air conditioning system; acquiring the air conditioning capacity of the air conditioning system; determining the priority of maintenance work for the air conditioning system in which signs of failure have been detected based on the air conditioning capacity; and outputting the priority along with the signs of failure.

[0020] According to the eleventh aspect of this disclosure, the accuracy of prioritizing maintenance work performed on air conditioners for which failures are predicted can be improved.

[0021] This diagram illustrates the flow of how maintenance services detect signs of failure. This is a scatter plot of an example of load factor and COP. This is a scatter plot of an example of load factor and COP. This is a flowchart diagram illustrating a general example of the process of outputting the priority of maintenance work when a failure is predicted. This is a diagram showing an example of the system configuration of a predictive detection system. This diagram illustrates an example of the arrangement of predictive detection devices and learning devices in a predictive detection system. This diagram illustrates an example of the arrangement of predictive detection devices and learning devices in a predictive detection system. This is a diagram illustrating an example of the arrangement of predictive detection devices and learning devices in a predictive detection system. This is a hardware configuration of an example of an edge device. This is a hardware configuration of an example of a server device. This is a functional block diagram illustrating an example of the functions of a predictive detection device in a predictive detection system divided into blocks. This is a functional block diagram illustrating an example of a learning device. This diagram illustrates a list of operating data included in the learning data. This is a diagram showing an example of learning data for generating a COP normal prediction model. This is a diagram showing an example of a COP normal prediction model constructed using a neural network. This is a flowchart illustrating an example of a learning method using a gradient boosting decision tree. This is a diagram illustrating the image of a decision tree. This is a diagram illustrating an example of training data to explain a gradient boosting decision tree. This is a diagram schematically illustrating the criteria for determining COP decline. This is a diagram illustrating a specific example of a threshold. This is a diagram illustrating a histogram when COP is normal. This is a diagram illustrating a histogram when COP is declining. This is a diagram illustrating an example of statistical data showing how long it takes from fault prediction to actual failure. This is a diagram illustrating an example of a notification screen that notifies that a failure has been predicted and the component it concerns. This is a diagram illustrating an example of a notification screen when no COP decline has been detected. This is a diagram illustrating an example of COP before and after maintenance work. This is a diagram illustrating an example of the measured COP and predicted COP for the past three months for a certain air conditioner. This is a diagram illustrating an example of the average COP for the past three months. This is a diagram illustrating a rapidly declining COP. This is a diagram illustrating a gradually declining COP. This is a diagram illustrating the correspondence between the rate of COP decline and an example of the timing and location of failure. This is a flowchart illustrating an example of a method for determining priority when a failure is predicted.This is an example flowchart illustrating the process performed by a predictive maintenance device when a systems engineer (SE) responds to a fault prediction.

[0022] Below, we will describe an example of a form for implementing this disclosure, including a predictive detection system and a predictive detection method performed by the predictive detection system.

[0023] <Overview of Fault Prediction and Service> When an air conditioning system (hereinafter referred to as "air conditioner") malfunctions, it becomes difficult to control the temperature of the air-conditioned space. This can cause discomfort if people are living in the space, and may cause deterioration of goods stored in spaces such as warehouses. While there are methods such as duplicating the air conditioning system, this increases costs. For this reason, the predictive maintenance system provides a maintenance service that remotely monitors air conditioners 24 hours a day, 365 days a year.

[0024] Figure 1 illustrates the process by which maintenance services detect signs of failure.

[0025] (1) For example, the outdoor unit 31 is connected to the edge device 10, which will be described later, and the outdoor unit 31 transmits operating data to the edge device 10 in short cycles of minutes. Here, we assume that the two outdoor units 31 are in a state where signs of failure are detected and a state where a failure is detected.

[0026] (2) The edge device 10 transmits the operation data to the server device 60.

[0027] (3) The server device 60 analyzes the operating data to detect signs of failure (hereinafter sometimes referred to as failure prediction) or to detect a failure (the outdoor unit 31 or edge device 10 may also detect it). If the server device 60 detects a failure, it notifies the service engineer (hereinafter referred to as SE9) that it has detected a failure. If the server device 60 predicts a failure, it notifies the SE9 that it has predicted a failure. The SE9 is a worker who performs operation, maintenance, and servicing of the installed equipment or service.

[0028] (4) In the event of a malfunction, SE9 will perform maintenance work with the goal of "responding immediately." If a malfunction is predicted, SE9 will aim to "perform maintenance work within a certain period." Through this operational policy for maintenance services, the maintenance service aims to prevent serious malfunctions and ensure that the air conditioners run without interruption.

[0029] However, securing SE9 personnel has become difficult, and a significant increase in SEs in line with the number of maintenance service contracts is not realistic. Therefore, an increase in the number of maintenance service contracts is expected to significantly increase the workload on the field, including SE9. For example, there may be cases where a failure is predicted, but the response within two weeks is not possible, leading to a failure. If a failure occurs, SE9 must perform maintenance work within two hours, further increasing the workload on the field.

[0030] While having SE9 perform maintenance on air conditioners with the highest urgency can reduce the number of cases leading to malfunctions, currently, appropriate prioritization of maintenance work based on fault prediction is not being carried out.

[0031] Therefore, the maintenance service of this embodiment aims to reduce the number of cases leading to failure and to reduce the workload on site by assigning appropriate priorities to maintenance work in response to failure prediction. The applicant focused on an index called COP (Coefficient of Performance), which represents the operating efficiency of the air conditioner, as a method for determining priority. COP is the air conditioning capacity per unit of power.

[0032] Figures 2A and 2B are scatter plots of load factor and COP. The COP and load factor in Figures 2A and 2B were calculated from operating data of the same model. Figures 2A and 2B show that there is a correlation between a decrease in COP and air conditioner failure. Figures 2A and 2B show COP 301 under normal conditions and COP 302 three months before failure, respectively. It can be seen that COP 302 three months before failure is decreasing compared to COP 301 under normal conditions. Therefore, it is thought that detecting a decrease in COP can improve the accuracy of prioritizing maintenance work.

[0033] However, since the COP varies significantly depending on the surrounding external environment and the usage of the air conditioner, as shown in FIGS. 2A and 2B, even for the same model of air conditioner, the distributions are significantly different. Also, the distributions of the COP302 three months before the failure and the COP301 during normal operation overlap, making it difficult to determine whether the COP has decreased (determination of the threshold value).

[0034] Therefore, in this embodiment, the learning device creates a regression model for predicting the COP during normal operation. When predicting a failure and using the decrease in COP to determine the priority of maintenance work, the predicted value of the COP in the normal state output by the regression model (an example of the expected air conditioning capacity, hereinafter referred to as the predicted COP) is compared with the measured COP (an example of the measured air conditioning capacity, hereinafter referred to as the measured COP) to determine the decrease in COP.

[0035] FIG. 3 is a schematic flowchart for explaining the process of determining the priority of maintenance work when a failure is predicted.

[0036] Although many components are used in the air conditioner, the server device 60 can repeatedly acquire the operation data output by each component. The server device 60 applies the operation data to the omen detection logic to detect the omen of failure for each component (S1). Note that the omen detection logic uses a rule - based approach, but an omen detection model may also be used.

[0037] Also, when a failure is predicted, the server device 60 calculates the COP (S3). The server device 60 has a COP normal judgment model described later. The COP normal judgment model is a model for predicting the COP during normal operation. The server device 60 determines whether the COP has decreased by comparing the predicted COP output by the COP normal judgment model with the measured COP calculated from the operation data.

[0038] Then, the server device 60 uses the COP to determine the priority of the maintenance work for the component for which the failure is predicted (S3). For example, when the COP has decreased, the server device 60 sets the priority of the maintenance work for the component for which the failure is predicted higher than when the COP has not decreased.

[0039] The server device 60 outputs basic information and priorities of air conditioners that require maintenance work (S4). For example, assume that a failure has been predicted for the air conditioner of customer A and its COP has decreased. Assume that a failure has been predicted for the air conditioner of customer B and its COP has not decreased. In this case, the server device 60 sets the priority of the maintenance work for the air conditioner of customer A higher than the priority of the maintenance work for the air conditioner of customer B.

[0040] By doing so, the SE can perform maintenance work starting from the air conditioners with high urgency, and can reduce the cases leading to failures. Since the SE can perform maintenance work according to the priorities, the on-site load can be reduced.

[0041] Note that the server device 60 can determine multiple levels of priorities according to the degree of decrease in COP, rather than determining the priority only based on whether the COP has decreased.

[0042] <Regarding Terms> A failure indicates a state in which there are some abnormalities, deteriorations, malfunctions, etc. in the device, resulting in a decrease in the performance and quality of the device. A failure means that it is highly likely that the air conditioner cannot be operated without maintenance work, or even if it can be operated, it cannot be operated early. Alternatively, a failure means that even if it is operated, it cannot perform its original function of controlling the temperature of the space. A failure may be called an "abnormality". In contrast, an omen of a failure means that it has not reached a failure but is different from the normal state.

[0043] Maintenance work means taking measures to return a failed or failure-predicted air conditioner to a state where it is not failed or failure-predicted. For example, it means performing work such as repair, maintenance, preservation, inspection, etc.

[0044] Air conditioning capacity is the amount of heat that an air conditioner can provide per unit time during cooling or heating. In this embodiment, the COP, which is the cooling or heating capacity per 1 [kW] of power consumption, is used as the air conditioning capacity.

[0045] A record refers to one row of data when data is arranged in a two-dimensional table. One column of data is a column. In this disclosure, different types of operational data are arranged in the row direction, and operational data is arranged chronologically in the column direction.

[0046] <System Configuration of the Predictive Detection System> Next, the system configuration of the predictive detection system will be explained with reference to Figure 4. Figure 4 is a diagram showing an example of the system configuration of the predictive detection system.

[0047] The predictive maintenance detection system provides a variety of IoT-based services to users, from administrators to general users, by enabling communication between various devices 30 and a cloud-based server device 60 via a network N. The edge devices 10, devices 30, sensor switches 20, and user terminals 5 are mainly located on the customer's side, while the server device 60 is located in a data center or the cloud via the internet. The user terminal 5 may be located at the maintenance service company to which the system engineer (SE) belongs, or it may be the SE's mobile terminal.

[0048] Equipment 30 refers to all devices that consume power, such as air conditioners 40, security equipment, heat source equipment, fire alarms, AHUs (air handling units), electricity meters, and lighting. Equipment 30 may also include other devices. Sensor switches 20 include various sensors, lamps, relays, etc. Equipment 30 and sensor switches 20 are connected to the edge device 10 so as to be able to communicate via a dedicated cable or a network such as a LAN. Equipment 30 and sensor switches 20 may also be connected to the edge device 10 so as to be able to communicate via wireless communication.

[0049] The equipment 30 and sensor switches 20 are controlled by the edge device 10. In other words, the edge device 10 performs the necessary operations on the equipment 30 and sensor switches 20 to suit the purpose of the equipment 30 and sensor switches 20. The content of the control varies depending on the type of equipment 30 and sensor switches 20, but for example, if the equipment 30 is an air conditioner, it may include all control related to the functions of the air conditioner, such as the cooling / heating mode, set temperature, airflow, humidity, and airflow direction, which can generally be set on an air conditioner. In addition, the control may include operating modes such as a pre-season inspection mode, microcontroller reset, operation stop, and function substitution.

[0050] Device 30 collects operational data corresponding to its own specifications and transmits it to the edge device 10 mainly on a regular basis. "Regularly" means, for example, once every minute, once every 10 minutes, once every 60 minutes, etc., but this can be set by the user or the server device 60. Furthermore, device 30 can transmit operational data to the edge device 10 upon request from the edge device 10 or the user terminal 5. The operational data varies depending on the device 30, but for example, in the case of an air conditioner, it may include outside air temperature, load factor, outdoor unit operating time, compressor rotation speed, refrigerant high pressure pressure, low pressure pressure, refrigerant temperature, fan rotation speed, and microcontroller CPU temperature.

[0051] Furthermore, if device 30 detects an abnormality, it sends an abnormality code to the edge device 10. The abnormality detected by device 30 is one that makes it difficult to continue operation. Device 30 that detects an abnormality stops operation. The edge device 10 sends an abnormality code to the server device 60. The processing of the edge device 10 for sensor switches 20 can be the same. Sensor switches 20 mainly periodically send information about themselves to the edge device 10 and send abnormality codes. The timing of transmission is periodically, at a set date and time, during periods of low load, when the difference from the last image data exceeds a certain amount, etc.

[0052] The edge device 10 is a controller that controls the equipment 30 and sensor switches 20. The edge device 10 functions as a control device that controls the equipment 30 and sensor switches 20, an information processing device that processes operating data, etc., and a communication means for communicating with the server device 60. For example, the edge device 10 transmits various information from the equipment 30 to the server device 60 and receives instructions from the server device 60 according to the information. Alternatively, the edge device 10 can receive instructions from the server device 60 even if it does not transmit any information to the server device 60 (for example, when there are instructions from the user terminal 5 to the server device 60). The edge device 10 converts the instructions into appropriate instructions according to the models of the equipment 30 and sensor switches 20 and transmits them to the equipment 30 and sensor switches 20. The equipment 30 or sensor switches 20 may be connected to the network N without going through the edge device 10.

[0053] A server device 60 is one or more information processing devices. Although Figure 4 shows one server device 60, the server devices 60 may be divided into several units according to their functions. Alternatively, the functions of the server devices 60 may be consolidated into a single information processing device. Furthermore, multiple server devices 60 with the same functions may be provided, and these multiple server devices 60 may communicate with each other to process data, like a server cluster.

[0054] The server device 60 receives various information transmitted from the edge device 10 via the network N and generates necessary instructions. In this embodiment, the server device 60 detects signs of failure (fault prediction) or detects a failure based on operating data transmitted from the equipment 30 or sensor switches 20. The detection of signs of failure or failure may be performed by the edge device 10 or the equipment 30. The server device 60 determines the priority of maintenance work for the air conditioner 40 in which signs of failure have been detected. The server device 60 transmits the priority of maintenance work for the air conditioner 40 in which signs of failure have been detected to the user terminal 5. In addition, in response to an abnormal code from the equipment 30 via the edge device 10 (in response to fault detection by the equipment 30), the server device 60 instructs the edge device 10 to perform emergency operation regardless of the model of the equipment 30. The server device 60 can also transmit instructions to the equipment 30 to the edge device 10 according to the schedule and operations set by the user terminal 5.

[0055] The server device 60 also has the functionality of a web server. The web server responds to requests from client software (web clients) such as a web browser operated by the user and provides the client with screen information written in HTML files, XML, CSS files, JavaScript (registered trademark), etc. An application that uses the web mechanism in this way is called a web application.

[0056] Furthermore, it is preferable that the server device 60 supports cloud computing. Cloud computing refers to a usage model in which network resources are utilized without the user being aware of specific hardware resources. Cloud computing provides users with data and software that they previously used on their own computers, as a service via the network. Users can access a variety of services from any device by providing a web browser that runs on a personal computer or mobile device, and an internet connection.

[0057] The user terminal 5 is a terminal device that displays various screens provided by the server device 60. In this embodiment, the user terminal 5 may be used by a system engineer, a customer's administrator, or a general user. That is, the user terminal 5 may be used by a customer's administrator or a general user. Alternatively, the user terminal 5 may be used by the administrator of the predictive maintenance system 100. The customer's administrator is the person who performs maintenance and management that is not performed by general users who use the equipment 30 on a daily basis.

[0058] The user terminal 5 displays a wide variety of screens, but one example is a screen that notifies users of information about an air conditioner 40 that has shown signs of malfunction and the priority of maintenance work. Other screens include a list of equipment 30 and sensor switches 20 connected to the customer's edge device 10, an in-house map showing the locations of the equipment 30 and sensor switches 20, and operation screens for operating the equipment 30 and sensor switches 20.

[0059] The user terminal 5 may be, for example, a PC (Personal Computer), smartphone, tablet, PDA (Personal Digital Assistant), or wearable PC (sunglasses type, wristwatch type, etc.). However, it only needs to have communication capabilities and be able to run a web browser. Alternatively, instead of a web browser, the user terminal 5 may run a dedicated native application for the predictive maintenance system 100.

[0060] <Examples of arrangement of learning device and predictive detection device> Figures 5A to 5D illustrate examples of the arrangement of the predictive detection device 400 and the learning device 500 in the predictive detection system 100. The learning device 500 is an information processing device that generates a COP normal judgment model during the learning phase of machine learning. The predictive detection device 400 is an information processing device that performs fault prediction detection, judgment of COP deterioration using the COP normal judgment model, and determination of maintenance work priority.

[0061] In Figure 5A, the learning device 500 is arranged to be connectable to the server device 60 and the edge device 10 via the network N. The predictive detection device 400 may be located on the air conditioner 40, the edge device 10, or the server device 60. The predictive detection device 400 may also be located on the network N independently of these. The functions of the predictive detection device 400 may be distributed across the air conditioner (the number 40, not shown in the figure), the edge device 10, or the server device 60.

[0062] In Figure 5B, the learning device 500 is integrated with the server device 60. That is, the server device 60 also functions as the learning device 500. The predictive detection device 400 may be located in the air conditioner 40, the edge device 10, or the server device 60. The predictive detection device 400 may also be located independently on the network N. The functions of the predictive detection device 400 may be distributed across the air conditioner 40, the edge device 10, or the server device 60.

[0063] In Figure 5C, the learning device 500 is integrated with the edge device 10. That is, the edge device 10 also functions as the learning device 500. The predictive detection device 400 may be located on the air conditioner 40, the edge device 10, or the server device 60. The predictive detection device 400 may also be located independently on the network N. The functions of the predictive detection device 400 may be distributed across the air conditioner 40, the edge device 10, or the server device 60.

[0064] In Figure 5D, the air conditioner 40 has a learning device 500 and a predictive detection device 400. That is, the air conditioner 40 serves as both the learning device 500 and the predictive detection device 400. The predictive detection device 400 may be arranged independently of the air conditioner 40.

[0065] In all configuration examples, during the learning phase, the learning device 500 generates a COP normal judgment model using learning data (training operation data) acquired from the air conditioner 40. During the inference phase, the predictive detection device 400 predicts failures of the air conditioner 40 by, for example, applying a rule-based approach to the operation data acquired from the air conditioner 40. The predictive detection device 400 also calculates the measured COP using the COP normal judgment model and determines the priority of maintenance work by judging a decrease in COP.

[0066] <Hardware Configuration of Edge Device and Server Device> Next, the hardware configuration of the edge device 10 will be described with reference to Figure 6. Figure 6 is a hardware configuration diagram of an example of the edge device 10. As shown in Figure 6, the edge device 10 has a processor 201, memory 202, auxiliary storage device 203, I / F (Interface) device 204, communication module 205, and drive device 206. Each piece of hardware in the edge device 10 is interconnected via a bus 207.

[0067] The processor 201 has various computing devices such as a CPU (Central Processing Unit). The processor 201 reads various programs into the memory 202 and executes them. The processor 201 corresponds to the control unit 110 that controls the entire edge device 10.

[0068] Memory 202 has main memory devices such as ROM (Read Only Memory) and RAM (Random Access Memory). The processor 201 and memory 202 form a so-called computer, and the processor 201 executes various programs read into memory 202.

[0069] The auxiliary storage device 203 stores various programs and various data used when those programs are executed by the processor 201.

[0070] The I / F device 204 is a connection device that connects the edge device 10 to an example of an external device, such as equipment 30 and sensor switches 20.

[0071] The communication module 205 is a communication device for communicating with the server device 60 via the network N.

[0072] The drive device 206 is a device for setting the recording medium 208. The recording medium 208 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 208 may also include semiconductor memory that records information electrically, such as ROMs and flash memory.

[0073] The various programs to be installed on the auxiliary storage device 203 are installed, for example, when the distributed recording medium 208 is set in the drive device 206 and the various programs recorded on the recording medium 208 are read by the drive device 206. Alternatively, the various programs to be installed on the auxiliary storage device 203 may be installed by downloading them from the network N via the communication module 205.

[0074] On the other hand, Figure 7 shows an example of the hardware configuration of the server device 60. Since the hardware configuration of the server device 60 is generally the same as that of the edge device 10, this explanation will focus on the differences between the two. Furthermore, the hardware configuration of the learning device, which will be described later, will be the same as that of the server device 60.

[0075] The processor 221 reads various programs into the memory 222 and executes them. The processor 221 corresponds to the control unit 210 that controls the entire server device 60.

[0076] The I / F device 224 is a connection device that connects the server device 60 to an external device, such as a display device 230 and an operating device 240. The display device 230 displays the internal status of the server device 60. The operating device 240 is used by the administrator of the server device 60 to input various instructions to the server device 60.

[0077] The communication module 225 is a communication device for communicating with the edge device 10 and the user terminal 5 via the network N.

[0078] <About the Functions> Next, with reference to Figure 8, the functional configuration of each device in the predictive detection system 100 will be explained in detail. Figure 8 is an example of a functional block diagram that explains the functions of the predictive detection device 400 in the predictive detection system 100 by dividing them into blocks. The predictive detection device 400, user terminal 5, and learning device 500 are connected to each other via the network N so as to be able to communicate. The user terminal 5 only needs to connect to the predictive detection device 400 when it needs to identify an air conditioner that has detected a predictive error, and may not be included in the predictive detection system 100. Also, the learning device 500 is not included in the predictive detection system 100 during the inference phase. The predictive detection device 400 may have its functions distributed and located on either the server device 60 or the edge device 10, or both.

[0079] <<Learning Device>> The learning device 500 generates a COP normality judgment model during the learning phase of machine learning. As described above, the learning device 500 can be any information processing device. Details of the functions of the learning device 500 will be described later.

[0080] <<Predictive Detection Device>> The predictive detection device 400 includes an operation data acquisition unit 41, a predictive detection unit 42, a COP calculation unit 43, a COP reduction judgment unit 44, a priority determination unit 45, a reception unit 46, and an output unit 47. Each of these parts of the predictive detection device is a function or means realized by any of the components shown in Figure 6 (or Figure 7) operating according to instructions from the processor 201 (or 221) in accordance with a program deployed from the auxiliary storage device 203 (or 223) to the memory 202 (or 222).

[0081] These components of the predictive detection device 400 are functions in the inference phase. The inference phase in this embodiment is a phase in which fault prediction is performed on the air conditioner 40 in operation. If a fault prediction is detected, the priority of maintenance work on the air conditioner 40 in which the fault prediction was detected is determined.

[0082] The operation data acquisition unit 41 repeatedly acquires operation data in real time from, for example, the air conditioner 40 while it is in operation. The acquisition period can be set according to the processing capacity of the predictive detection device 400 and the network bandwidth, for example, every minute. Furthermore, the acquisition period does not necessarily have to be regular.

[0083] The predictive detection unit 42 uses predictive detection logic (rule-based) to detect signs of failure in the air conditioner 40. When using a rule-based approach, the predictive detection unit 42 compares operating data with one or more thresholds, and if the result is above or below the threshold, it predicts a failure. Alternatively, the predictive detection unit 42 may use a predictive detection model to detect signs of failure. In this case, the predictive detection unit 42 inputs operating data into the predictive detection model, calculates predicted values ​​for normal operation, and determines whether the deviation between the measured values ​​and predicted values ​​included in the operating data is above a threshold. If the deviation is above a threshold, the predictive detection unit 42 predicts a failure.

[0084] The COP calculation unit 43 calculates a predicted COP using a COP normal judgment model when a fault is predicted, and calculates the actual COP using the operating data. The COP calculation unit 43 calculates the actual COP by dividing the cooling (heating) capacity calculated using the operating data by the power consumption.

[0085] The COP reduction determination unit 44 determines whether the measured COP has decreased compared to the predicted COP, based on the COP reduction determination criteria described later. If the measured COP has decreased compared to the predicted COP, it means that the likelihood of the fault-predicted component failing has increased.

[0086] The priority determination unit 45 determines the priority of maintenance work based on whether the measured COP has decreased relative to the predicted COP when a failure is predicted. In other words, the priority determination unit 45 sets a higher priority for maintenance work on an air conditioner 40 that has been predicted to fail and whose measured COP has decreased relative to the predicted COP, than for maintenance work on an air conditioner 40 whose measured COP has not decreased relative to the predicted COP.

[0087] The reception unit 46 receives confirmation from the user terminal 5 that the maintenance work has been completed when the SE (System Engineer) inputs this information to the user terminal 5. The reception unit 46 records the date and time the maintenance work was completed and is able to inform the SE or the customer of how much the COP (Coefficient of Performance) improved before and after the maintenance work.

[0088] The output unit 47 outputs instructions for maintenance work with assigned priorities to the user terminal 5. The user terminal 5 may be a terminal device that the SE can access, or it may be a customer-side terminal device (in this case, the customer requests maintenance work from the SE). The output unit 47 may send the maintenance work instructions via email or via social networking service (SNS). The output unit 47 may also output the maintenance work instructions to a website displayed on the user terminal 5. Alternatively, the output unit 47 may push notifications of maintenance work instructions to applications running on the user terminal 5.

[0089] <About the functions of the learning device> Next, with reference to Figure 9, the functions of the learning device 500 will be described. Figure 9 is a functional block diagram of the learning device 500 according to one embodiment of the present disclosure. The learning device 500 generates a COP normal prediction model 12.

[0090] The learning device 500 includes a learning data acquisition unit 502, a learning data storage unit 503, and a learning unit 504. These functional units of the learning device 500 are functions or means realized by the processor 221 of the learning device 500 executing program instructions that have been loaded into the memory 222.

[0091] The learning data acquisition unit 502 acquires learning data 501 and stores it in the learning data storage unit 503. The learning data 501 consists of one or more operating data points under normal conditions and further comprises multiple records. "Normal" means that there are no abnormalities in all thermistors, actuators, compressors, and expansion valves of the air conditioner 40 when it provides its function. Each record contains explanatory variables and objective variables for generating the COP normal prediction model 12. Details of these will be described later.

[0092] Furthermore, it is preferable for the learning data acquisition unit 502 to perform preprocessing on the learning data 501. Preprocessing includes, for example, removing transient data and operating data to which error codes are attached. Transient data refers to operating data acquired in a transient state where the room temperature of the space to be air-conditioned is not in a steady state.

[0093] The learning data storage unit 503 stores the operation data acquired by the learning data acquisition unit 502 and after preprocessing has been completed.

[0094] The learning unit 504 learns from training data using various machine learning algorithms to generate a COP normal prediction model 12. The COP normal prediction model 12 is correspondence information that maps driving data corresponding to explanatory variables to COP. The COP normal prediction model 12 outputs a predicted COP for input driving data corresponding to explanatory variables. Such correspondence information can be realized by a regression model. Regression models include multiple regression, gradient boosting decision trees, AdaBoost, neural networks, ridge regression, Lasso regression, and elastic network regression.

[0095] Note that the driving data includes multiple types of data, and all of them can potentially serve as the target variable.

[0096] <About the Driving Data> Refer to Figures 10A and 10B to explain the driving data in the training data. Figure 10A shows a list of the driving data included in the training data.

[0097] As an example, let's assume that operating data A to Z are acquired from the air conditioner 40. Operating data A1, A2, ... are operating data A acquired in chronological order. A set of operating data collected from the air conditioner 40 at the same time is called a record. In Figure 10A, operating data A1 to Z1 constitute one record, operating data A2 to Z2 constitute one record, ..., operating data A7 to Z7 constitute one record. In other words, one record contains multiple types of operating data. These include, for example, outside air temperature, compressor rotation speed, load factor, outdoor unit operating time, power consumption, and cooling / heating flags. In this embodiment, data calculated from operating data using formulas, such as COP and load factor, are also referred to as operating data.

[0098] FIG. 10B is an example of learning data for generating the COP normal prediction model 12. This learning data is included in the operation data as shown in FIG. 10A or is calculated based on the operation data. The target variable of the COP normal prediction model 12 is the COP. The explanatory variables may be operation data suitable for explaining the COP among the operation data A to Z. For example, they may be the outside air temperature, the compressor rotation speed, the load factor, the outdoor unit operation time, the cooling / heating flag, etc. shown in FIG. 10B. Other operation data may be included in the explanatory variables of the COP normal prediction model 12, or some explanatory variables may not be used.

[0099] The COP and the load factor are calculated by the following formulas. COP = Cooling (or heating) capacity [kW] / Power consumption [kW] Load factor = Cooling (or heating) capacity [kW] / Rated capacity [kW] The COP and the load factor may be included in the operation data transmitted by the air conditioner 40, or may be calculated by the sign detection device 400.

[0100] Although the calculation formulas for the COP and the load factor are different for cooling operation and heating operation respectively, since the cooling / heating flag is included in the explanatory variables, the COP normal prediction model 12 is common for cooling operation and heating operation. The COP normal prediction models 12 for cooling operation and heating operation may be generated separately. The number of records of the learning data preferably includes the amount for one year or more from the start of acquisition, and is 1500 or more for each of the cooling operation and the heating operation. Also, the COP normal prediction model 12 is generated for each same system (the same outdoor unit).

[0101] <Regression model> The COP normal prediction model 12 using the regression model will be described.

[0102] For example, in the case of multiple regression, the target variable and the explanatory variables are associated as follows. y = β ０ +β １ x １ +β ２ x ２ +……+β ｎ x ｎ　　 ……(1) y is the target variable, x １ ~x ｎis an explanatory variable. In the case of the COP normal prediction model 12, the dependent variable is COP, and the explanatory variables are outside air temperature, compressor rotation speed, load factor, outdoor unit operating time, power consumption, and cooling / heating flags. β ０ ~β ｎ is the coefficient of the explanatory variable. Multiple regression analysis is performed on β ０ ~β ｎ This is determined. The model shown in equation (1) above is the COP normal prediction model 12. Note that a predictive detection model can be generated in the same way.

[0103] As shown in Figure 11, the learning device 500 may generate a COP normal prediction model 12 using a neural network. Figure 11 shows a COP normal prediction model 12 configured with a neural network 170. The neural network 170 is a form of artificial intelligence (AI) model that trains a computer to process data in a way that mimics the workings of the human brain. An existing configuration will be used for the neural network 170.

[0104] In Figure 11, the inputs to the neural network are the outside temperature, compressor rotation speed, load factor, outdoor unit operating time, power consumption, cooling / heating flag, and bias (set to 1). The output is the target variable (e.g., the predicted value of COP). The COP normal prediction model 12 has an input layer 171, a hidden layer 172, and an output layer 173. In the neural network 170 of Figure 11, three layers (excluding the input layer) are fully connected from the nodes of the input layer 171 to the nodes of the output layer 173. A neural network with deep layers is called a DNN (Deep Neural Network). The layer between the input layer 171 and the output layer 173 is called the hidden layer 172. The number of hidden layers 172 and the number of nodes are merely examples.

[0105] Weights are assigned to the connections between nodes, and the output from a node multiplied by its weight is transmitted to the node in the next layer. The node in the next layer receives the outputs of all the nodes in the previous layer, so it sums up the outputs of all the nodes in the previous layer. The node in the next layer activates the summed output with an activation function and transmits it to the next node. This process is repeated until the values ​​are transmitted up to output layer 173.

[0106] In this embodiment, since we want to infer the predicted value of the target variable, the COP normal prediction model 12 is a regression model (a classification model is another option). For this reason, the output layer 173 is provided with one node that outputs the predicted value of the target variable.

[0107] It should be noted that the learning phase of the COP normal prediction model 12 was performed using existing methods such as backpropagation. The COP normal prediction model 12 learns the correspondence between explanatory variables (outside air temperature, compressor rotation speed, load factor, outdoor unit operating time, power consumption, and cooling / heating flags) and the target variable (COP).

[0108] In the inference phase using the COP normal prediction model 12, for example, current operating data (such as outside air temperature) transmitted from the operating air conditioner 40 is input to the input layer 171. The output layer 173 calculates (infers) the predicted value of COP. By repeatedly outputting such predicted values, predicted values ​​associated with time can be obtained.

[0109] The neural network structure can be the same for predictive maintenance models. In predictive maintenance models, the input to the input layer is operational data B, C, D, etc., and the output is the predicted value of operational data A.

[0110] Next, we will explain the regression model using gradient boosting decision trees, referring to Figures 12 to 14. Figure 12 is a flowchart illustrating the learning method using gradient boosting decision trees. Figure 13 shows an image of the decision tree. Figure 14 is the training data used to explain the gradient boosting decision tree. In Figure 14, the target variable is, for example, COP, and explanatory variables 1 and 2 are the outside air temperature and compressor rotation speed. Other explanatory variables (load factor, outdoor unit operating time, power consumption, cooling / heating flag) have been omitted. Also, the numerical values ​​in Figure 14 are for illustrative purposes only and differ from actual values.

[0111] The learning unit 504 calculates initial values ​​from the values ​​of multiple target variables (e.g., COP) prepared as training data (S11). The initial values ​​are, for example, the mean values. For the sake of explanation, this mean value is called the predicted value 1.

[0112] Next, the learning unit 504 calculates the error 1 for each record of the training data by subtracting the mean value from the value of the target variable (S12). Error 1 is positive if the value of the target variable is greater than the mean, and negative if it is smaller. The reason why "the value obtained by subtracting the mean value from the value of the target variable" is used as the error is because the loss function is the squared error. The gradient obtained by differentiating the squared error is taken as the error. The absolute value error may also be used as error 1.

[0113] Next, the learning unit 504 constructs a decision tree for the purpose of predicting error 1 (S13). In other words, it classifies error 1 using a decision tree. The number of nodes and hierarchy of the decision tree may be limited to a predetermined range. This decision tree is a weak classifier (boosting). The top tree in Figure 13 is an image of the decision tree that is initially created. Figure 13 is an image, but error 1 is classified into the leaves (ends of the tree) of such a decision tree. In Figure 13, errors 1 (-35, -25, -15, -5, 5, 15, 25, 35) are classified into leaves. That is, a threshold is set for the nodes of the decision tree, and error 1 is classified into leaves by comparing whether explanatory variable 1 or 2 is greater than or less than the threshold for each node. If multiple errors are classified into one leaf, the average of the errors contained in that leaf is the error of that leaf.

[0114] A brief explanation of the decision tree construction is given below. (i) The learning unit 504 performs the following processing in order from the node closest to the root. First, the learning unit 504 calculates the entropy of the training data before classification. The learning unit 504 may use the Gini coefficient instead of entropy. (ii) The learning unit 504 sorts the training data by the value of an arbitrary explanatory variable. The learning unit 504 calculates the entropy of the training data classified into the two target branches, while changing the threshold for dividing the values ​​of the explanatory variable. The learning unit 504 performs this process for all attributes (explanatory variables 1 and 2). (iii) The learning unit 504 calculates the weighted average of the entropies from (ii). This weighted average may be the ratio of the number of data classified into each branch to the number of data before classification. (iv) The learning unit 504 adopts the attribute that yields the largest difference (gain) between the entropies from (i) and (iii) as the root attribute, and determines the threshold for each node. Therefore, the attribute and threshold that yield the greatest gain are determined as the classification conditions for each node. (v) The structure of the lower part of the branch to be classified is also determined by the processes in (i) to (iv).

[0115] Next, the learning unit 504 calculates a new predicted value using error 1 (S14). Since error 1 is classified for each sample of the training data in the leaves of the decision tree, a predicted value 2 can be calculated using error 1 for each sample of the training data. Note that although we used "error 1" here, "error n" increases with each iteration.

[0116] If we denote the new predicted value as predicted value 2, then "predicted value 2 = predicted value 1 + learning rate * error 1". The learning rate is a value less than 1 and is a hyperparameter that determines how much the error is corrected in a single decision tree iteration. The learning rate is determined appropriately, for example, between 0.05 and 0.3. In Figure 14, the learning rate is set to 0.1. By repeatedly performing the process of "finding the error, multiplying it by the learning rate, and adding it up," the accuracy gradually improves.

[0117] Next, the learning unit 504 calculates the error from the value of the target variable and the predicted value 2 (S15). The learning unit 504 calculates the error between the value of the target variable and the predicted value 2 from all the prepared training data. The error calculated from the predicted value 2 is called error 2.

[0118] The learning unit 504 repeats steps S13 to S15 (S16) until it has calculated the error a certain number of times or until the error falls below a threshold. In the second tree in Figure 13, error 2 is classified. The learning unit 504 classifies the previous error using a new decision tree and calculates predicted values ​​3, 4...n and errors 3, 4...n using the errors. Therefore, in the third tree, error 3 is classified. As the error changes with each repetition of the process, the structure of the decision trees also changes automatically. The n decision trees created in this way constitute a regression model using gradient boosting decision trees.

[0119] Various frameworks for implementing gradient boosting decision trees in information processing devices include LightGBM (Light Gradient Boosting Machine), XGBoost (eXtreme Gradient Boosting), and Catboost (Category Boosting). The learning unit 504 may use one of these frameworks.

[0120] This section describes inference using gradient boosting decision trees. The COP calculation unit 43 inputs operating data corresponding to explanatory variables obtained from the operating air conditioner 40 to all decision trees created in the learning phase. Therefore, the operating data is classified into one of the leaves according to the classification result of explanatory variable 1 or 2 based on the threshold set in the node. Since explanatory variables classified into the same leaf can be assumed to have similar target variables, the predicted value can be inferred from the error of the classified leaf. In Figure 13, there are three decision trees, and the operating data is classified into one leaf of each decision tree based on the value of explanatory variable 1 or 2. Each leaf stores the error with the predicted value. For example, suppose the data is classified as operating data in leaves 321 to 323, indicated by the dotted circles in Figure 13. The COP calculation unit 43 estimates the predicted value of the target variable (COP) by summing the values ​​obtained by multiplying the error of these leaves by the learning rate against the average value calculated in the learning phase across all decision trees.

[0121] Predicted value of the target variable = Mean + Learning rate × Error of leaf 321 + Learning rate × Error of leaf 322 + Learning rate × Error of leaf 323 = 135 + 0.1 × (-5) + 0.1 × (-22.5) + 0.1 × (32) Thus, in a gradient boosting decision tree, the final predicted value is the value obtained by multiplying the learning rate by the error of the target leaf in each decision tree and adding it to the mean calculated in the learning phase.

[0122] <Method for detecting signs of failure> Next, we will explain the method for detecting signs of failure. In the case of failure prediction using a rule base, the sign detection unit 42 compares the operating data with a threshold and detects signs of failure if the data is above or below the threshold. Signs of failure using a sign detection model are judged, for example, based on the degree of deviation between the predicted value output by the sign detection model and the measured value. The method for calculating the degree of deviation is as follows: Degree of deviation = predicted value - measured value Degree of deviation = (predicted value - measured value) / predicted value Degree of deviation = measured value / predicted value By quantifying the degree of deviation between the predicted value and the measured value in this way, the difference between the two can be obtained. The sign detection unit 42 detects signs of failure when the degree of deviation exceeds the threshold. The target of the detection of signs of failure is the component that outputs a measured value that deviates from the predicted value.

[0123] <COP Decrease Judgment Criteria> Next, the COP decline judgment criteria will be explained with reference to Figures 15A and 15B. Figure 15A is a schematic diagram illustrating the COP decline judgment criteria. Figure 15B is a specific example of a threshold. In Figure 15A, the horizontal axis represents the judgment period, and the vertical axis shows the predicted COP 18 and the measured COP 19. First, the terms used in the conditions of the COP decline judgment criteria will be explained. (1) Judgment window 15 The judgment window 15 is a period obtained by dividing the judgment period (e.g., 3 months) into one-week intervals, and is the unit for evaluating the decline in COP. The judgment window 15 is, for example, N1 [days]. (2) Decrease criterion 16 The decline criterion 16 is a reference value regarding how much the measured COP 19 has decreased from the predicted COP 18. The decline criterion 16 is expressed, for example, as σ (standard deviation) × r (r is a real number). The judgment criterion is, for example, N2 × σ (N2 times the standard deviation). Details are explained in Figures 16A and 16B. (3) Reduction rate 17 The reduction rate 17 is the percentage of data that meet the reduction criteria 16 within the judgment window 15. The reduction rate 17 is, for example, N3 [%].

[0124] When a fault is predicted, the COP reduction determination unit 44 acquires operating data from past determination periods and makes a COP reduction determination. The COP reduction determination unit 44 determines that the COP has decreased if there is at least one (1) determination window frame 15 that satisfies (2) reduction criteria 16 and (3) reduction ratio 17. Conversely, if there are no (1) determination window frames 15 that satisfy (2) reduction criteria 16 and (3) reduction ratio 17, it determines that the COP is normal.

[0125] By providing multiple COP reduction judgment criteria as shown in Figures 15A and 15B, the COP reduction judgment unit 44 can also determine the degree to which the COP has decreased. The priority determination unit 45 may determine multiple levels of priority according to the degree to which the COP has decreased. The COP reduction judgment criteria can be any predetermined judgment criteria that can determine a decrease in COP, and are not limited to those shown in Figures 15A and 15B.

[0126] The reduction criterion 16 will be explained with reference to Figures 16A and 16B. As described above, COP can change significantly depending on the surrounding external environment and how the air conditioner is used, so the measured COP 19 can fluctuate even under normal conditions. For this reason, it is preferable to understand how much the measured COP 19 fluctuates and to determine that the measured COP 19 has decreased compared to the predicted COP 18 when the measured COP 19 has decreased significantly more than under normal conditions.

[0127] Figures 16A and 16B show histograms of "predicted COP - measured COP". Figure 16A is the histogram when COP is normal, and Figure 16B is the histogram when COP is declining. Comparing the two, the histogram when COP is declining is larger than the histogram when COP is normal in the range of "0.2 to 0.7" between predicted and measured values. Therefore, it is conceivable to use "0.2 to 0.7" as the criterion for decline 16, but since "0.2 to 0.7" is a wide range of values, it is difficult to adopt it as a criterion value for measured COP 19, which fluctuates even under normal conditions. Therefore, we will use r, expressed as the standard deviation σ × r, as the criterion for decline 16.

[0128] This standard deviation is the standard deviation of the predicted COP output when the COP normal prediction model 12 is given the same explanatory variables as the training data. In other words, it is the standard deviation of the COP normal prediction model 12. The COP reduction judgment unit 44 gives the COP normal prediction model 12 the same explanatory variables as the training data, calculates "predicted COP - measured COP", and calculates the standard deviation σ of the calculation result. When operating data is acquired from the air conditioner 40 in operation, it is determined that (2) is satisfied if the measured COP 19 is smaller than the predicted COP 18 output by the COP normal prediction model 12 by a standard deviation of σ × N² or more.

[0129] <Determining the Priority of Maintenance Work> Next, we will explain how to determine the priority of maintenance work based on the results of the COP reduction assessment when a failure is predicted.

[0130] Figure 17 shows statistical data on the time elapsed from component failure prediction to actual failure. The horizontal axis represents the time elapsed from failure prediction to actual failure: within one week, within two weeks, within one month, within three months, and longer. The left vertical axis represents the number of failures, and the right vertical axis represents the cumulative percentage of failures.

[0131] The horizontal axis shows, in bar graph form, the number of actual failures (number of air conditioners) that occurred after failure prediction during the period. The bar graph on the left, 21, shows the number of actual failures when the COP was low, and the bar graph on the right, 22, shows the number of actual failures when the COP was normal. Comparing the bar graphs on the left, 21, and right, 22, it can be seen that air conditioners 40 with a low COP have a higher number of failures after failure prediction.

[0132] Figure 17 shows the cumulative percentage 23 when COP is low and the cumulative percentage 24 when it is normal. Cumulative percentages 23 and 24 are the cumulative values ​​of the occurrence rate of failures that occur by the relevant period. The occurrence rate is calculated as "number of air conditioners that actually failed / number of air conditioners that were predicted to fail". The cumulative percentage 23 when COP is low is consistently higher than the cumulative percentage 24 when it is normal. From this, it can be seen that the time from failure prediction to actual failure is shorter for air conditioners 40 with a low COP.

[0133] Therefore, if the COP normality prediction model 12 determines that the COP is decreasing, it is effective to prioritize maintenance work. For example, if a failure is predicted and the COP is determined to be decreasing, the output unit 47 outputs a message to the user terminal 5 suggesting maintenance work for the air conditioner. For example, to a system engineer, the message would be: "Signs of a failure have been detected in the air conditioner. The COP is also decreasing, so the priority is 'medium'. Please come and perform maintenance work within two weeks." For example, to a customer, the message would be: "Signs of a failure have been detected in the air conditioner. The COP is also decreasing, so the priority is 'medium'. Please request maintenance work from a service engineer."

[0134] More preferably, when a failure is predicted and a COP (Coefficient of Performance) decrease is detected, the output unit 47 outputs the following information to the SE: "Statistical data shows that X percent of air conditioners will fail within X months." Looking at the cumulative percentage 23 in Figure 17, approximately 40 percent of the air conditioners 40 fail within one month, so for example, information such as "Statistical data shows that 40 percent of air conditioners 40 will fail within one month" is sent to the user terminal 5. In addition, information such as "80 percent of air conditioners 40 will fail within one month" may be output, depending on the statistical data of the parts whose failures have been predicted. In this way, the output unit 47 can output the predicted timing of failure and the probability of failure occurring.

[0135] Statistical data like that shown in Figure 17 is obtained for each component of air conditioners in the same system (same outdoor unit), so the output unit 47 can create instructions for each component based on the statistical data. In addition, the information output to the user terminal 5 may only be the predicted time or the probability of failure occurring, as shown below: "Statistical data indicates that a failure will occur within ○ months" "Statistical data indicates that × percent of air conditioners will fail" Figures 18A and 18B illustrate the notification screen 330 that notifies the user that a failure has been predicted and the component it is referring to. By connecting the user terminal 5 to the predictive detection device 400, the notification screen 330 as shown in Figures 18A and 18B can be displayed. The output unit 48 generates the notification screen 330 and displays it on the user terminal 5. In the notification screen 330, the column direction represents the type of component, and the row direction represents whether a failure has been predicted, the 80% failure period, the latest prediction date and time, and the priority. Figure 18A is the notification screen 330 when a decrease in COP is detected.

[0136] - The "Fault Prediction Status" indicator shows whether the predictive detection unit 42 has predicted a fault. "None" indicates no fault prediction, and "×" indicates fault prediction (indicating that a fault has been predicted). "△" indicates that the predictive detection unit 42 has predicted a fault, but the number of predictions is below a specified value.

[0137] The 80% failure period is the time it takes for 80% of the predicted failure air conditioners 40 to fail. As described above, the 80% failure period is determined from statistical data (cumulative percentage 23). Figure 18A shows that the 80% failure period for thermistor failure prediction is "1 month". Also, the "8" in "80%" may differ depending on the component. In this case, the column for the 80% failure period will display information such as "Statistical data shows that × percent of air conditioners will fail within ○ months".

[0138] The latest predicted date and time is the date and time when the failure was last predicted. The first predicted date and time may be displayed separately from the latest predicted date and time, or together with the latest predicted date and time.

[0139] - The priority displayed is the priority determined by the priority determination unit 45. Since the COP is low, the priority is "medium". The priority may be displayed in three stages, such as high, medium, and low, or in five stages, such as 1 to 5.

[0140] Figure 18B shows the notification screen 330 when no COP decrease is detected (normal). The screen configuration is the same as in Figure 18B, but the 80% failure period is "3 months" and the priority is "low". "3 months" is determined by the cumulative ratio 24 of statistical data. A priority of "low" is sufficient if the priority is lower than the priority when COP is decreasing. For example, the priority when COP is decreasing may be "high" and the priority when COP is not decreasing may be "medium".

[0141] The predictive maintenance device 400 of this disclosure, when a failure is predicted in an air conditioner 40 and the COP (Coefficient of Performance) is low, prioritizes maintenance work on the air conditioner 40 higher than maintenance work on an air conditioner 40 where the COP is not low, thereby improving the accuracy of prioritizing maintenance work on an air conditioner 40 where a failure has been predicted. The system engineer (SE) can determine the failure response schedule by looking at the 80% failure period, the latest predicted date and time, and the priority for each component.

[0142] <Presenting the effects of maintenance work> When a failure is predicted, the air conditioner continues to operate. Therefore, if a system engineer (SE) performs maintenance work when a failure is predicted, the air conditioner 40 will continue to operate as before, making it difficult for the customer to understand what changes have occurred as a result of the SE's work.

[0143] Therefore, if the COP was low during the fault prediction stage, the output unit 47 could compare the COP before and after the maintenance work and present it to the customer after the maintenance work is completed. Since the COP, which was on a downward trend, returns to a normal value after the SE repair, the customer can confirm that the SE repair was indeed performed and that the COP has improved by comparing the COP before and after the repair.

[0144] Figure 19 shows an example of a COP before and after maintenance work. In Figure 19, the horizontal axis represents the date and time, and the vertical axis represents the COP. A system engineer (SE) performed maintenance work on the air conditioner 40 on date and time A. The SE inputs a message to the user terminal 5 indicating that the maintenance work is complete, and the user terminal 5 transmits this message to the predictive maintenance device 400. The receiving unit 46 of the predictive maintenance device 400 records the parts that have been maintained, associating them with the date and time.

[0145] The maintenance work targets a component for which failure has been predicted, for example, the cleaning of an external heat exchanger. Up until date and time A, the predicted COP 18 and the measured COP 19 deviate significantly, but after date and time A, the deviation becomes smaller. In this way, the output unit 47 can demonstrate to the customer that the repair was carried out appropriately, thereby improving customer satisfaction with the maintenance service.

[0146] Furthermore, instead of outputting the predicted COP 18 and measured COP 19 before and after the repair, the output unit 47 may output information regarding the air conditioning capacity of the air conditioner. For example, the output unit 47 may display a message such as "The COP of case XX has recovered to a normal level" or send a notification via email. Such a message can also show the customer that the repair was carried out properly, thereby improving customer satisfaction with the maintenance service.

[0147] <Presenting the decline in COP due to equipment degradation> Interest in energy conservation has been increasing for some time, and there is a need to reduce electricity costs. Currently, the measures that customers can take to reduce electricity costs are measures that may affect comfort, such as changing the set temperature.

[0148] For example, if the COP (Coefficient of Performance) is low, electricity costs can be saved by restoring it to normal through maintenance work. Therefore, it is considered effective to inform customers that electricity costs will be reduced by restoring the air conditioner 40 to a normal state. When the COP is low, the output unit 47 can visualize how much wasted electricity is being generated compared to a normal state, thereby encouraging repairs and saving energy.

[0149] Figure 20 shows an example of the measured COP 19 and predicted COP 18 for a certain air conditioner 40 over the past three months. In Figure 20, the horizontal axis represents the date and time, and the vertical axis represents the COP. The discrepancy between the predicted COP 18 and the measured COP 19 begins around date and time B. Therefore, it is considered that the COP decrease detection unit 44 detected a decrease in COP around date and time B. Furthermore, it is assumed that a failure is predicted in the not-too-distant period before or after the time when the COP decrease detection unit 44 detected a decrease in COP.

[0150] The COP calculation unit 43 calculates, for example, the average of the predicted COP and the measured COP over the past three months. The average of the past three months is shown in Figure 21. As an example, the average of the predicted COP 18 is R2, and the average of the measured COP 19 is R1 (R1 < R2).

[0151] Since the cooling (heating) capacity and power consumption are known for calculating the measured COP19, the power consumption is also known. The output unit 47 can estimate the electricity cost by multiplying the power consumption by the electricity unit price and the period. If the electricity cost related to the measured COP19 is denoted as electricity cost P, then the electricity cost Q related to the predicted COP18 is calculated as follows: Q = (R1 / R2) × electricity cost P. In this way, the output unit 47 can encourage maintenance by showing the customer that electricity costs will be reduced by restoring the air conditioner 40 to a normal state through maintenance work. The customer can also make a numerical decision on whether or not to request maintenance.

[0152] <COP Decline Rate> It is known that the nature of the failure and the time until the failure occurs vary depending on whether the COP decreases rapidly or gradually. Therefore, the COP decline determination unit 44 can determine the priority of maintenance work by determining the rate of COP decline.

[0153] Figures 22A and 22B show comparative examples of COP decline rates. Figure 22A shows a rapidly declining COP, while Figure 22B shows a gradually declining COP. In both cases, the predicted COP 18 shows little fluctuation, but the measured COP 19 is declining, and in Figure 22A, the slope is steeper. Considering the need for early fault response, the period for determining the decline rate can be set to approximately N1 [days], the same as the judgment window frame 15.

[0154] Figure 23 shows the correspondence between the COP degradation rate, failure timing, and failure location. Statistically, a higher COP degradation rate is associated with a shorter time to failure. Furthermore, statistically, a high COP degradation rate is often associated with failures in the refrigerant system (e.g., refrigerant leaks). Components that are highly likely to fail when the COP degradation rate is high are pre-configured in the predictive detection device 400.

[0155] The priority determination unit 45, if the component for which failure is predicted is a component that is particularly likely to fail when the COP decreases, sets the priority of the maintenance work higher than when a decrease in COP is simply detected.

[0156] <Operation Procedure> Figure 24 is a flowchart illustrating the method for determining priority when a failure is predicted. The process in Figure 24 may be executed each time the predictive detection device 400 acquires operating data from the air conditioner 40 that is in operation. Note that the method for determining priority in Figure 24 is merely an example and may vary depending on various factors such as parts, the degree of COP degradation, and the property.

[0157] First, when the operation data acquisition unit 41 acquires operation data, the predictive detection unit 42 uses predictive detection logic to detect signs of failure. Failure prediction is sufficient if it is detected using either a rule-based or predictive detection model. The predictive detection unit 42 determines whether or not a failure has been predicted for each component (S21). If no failure is predicted (No. in S21), the process in Figure 24 ends.

[0158] If a fault is predicted (Yes in S21), the COP calculation unit 43 calculates the predicted COP 18 and the measured COP 19 (S22).

[0159] The COP reduction determination unit 44 determines whether the measured COP 19 has decreased based on the COP reduction determination criteria (S23).

[0160] If the measured COP19 has not decreased (No. in S23), the priority determination unit 45 determines that the priority is "low" (S27).

[0161] If the measured COP 19 is decreasing (Yes in S23), the COP decrease determination unit 44 determines whether the rate of decrease of the measured COP 19 is above a threshold and whether the faulty component is part of the refrigerant system (S24). The faulty component may be a component other than the refrigerant system if it is a component that is prone to failure when the COP decreases. Alternatively, in step S24, it may be determined only whether the rate of decrease of the measured COP 19 is above a threshold, without determining whether the component is prone to failure when the COP decreases.

[0162] If the decision in step S24 is Yes, the priority determination unit 45 determines that the priority is "High" (S25). If the decision in step S24 is No, the priority determination unit 45 determines that the priority is "Medium" (S26).

[0163] Figure 25 is a flowchart illustrating the process performed by the predictive detection device 400 when a system engineer (SE) responds to a fault prediction. Note that the flow in Figure 25 is just one example and may vary depending on various factors such as the nature of the fault and customer requirements.

[0164] If a malfunction is predicted, the output unit 47 notifies the SE by sending an email to the user terminal 5 containing the URL of the predictive detection device 400. When the user terminal 5 connects to the predictive detection device 400, the output unit 47 causes the user terminal 5 to display a notification screen 330 as shown in Figures 18A and 18B (S31).

[0165] The SE looks at the notification screen 330, determines which air conditioners 40 and their parts have high priority, and performs maintenance work. If an individual contract is required for the maintenance work, the SE should present the customer with a graph comparing the measured COP 19 and predicted COP 18, as shown in Figure 20, or the average value of the measured COP 19 and predicted COP 18 (the electricity cost may also be presented), as shown in Figure 21. In this case, the SE inputs an operation to display the information in Figures 20 and 21 into the user terminal 5, and the user terminal 5 obtains this information from the predictive maintenance device 400.

[0166] When the SE completes the maintenance work, it inputs a message to the user terminal 5 indicating that the maintenance work has been completed. The user terminal 5 transmits the parts that were maintained and the fact that the maintenance work has been completed to the predictive maintenance device 400. The receiving unit 46 of the predictive maintenance device 400 associates the maintained parts with the date and time and saves the message that the maintenance work has been completed (S32).

[0167] For example, if a certain period of time (e.g., 3 months) has elapsed since maintenance work, the COP calculation unit 43 calculates the measured COP 19 and predicted COP 18 after the maintenance work based on the operating data after the maintenance work. The COP reduction determination unit 44 determines whether the COP has improved or not (S33). In other words, it determines whether the measured COP 19 after the maintenance work has reached a state where it is not judged to have decreased, based on the COP reduction determination criteria.

[0168] If the determination in step S33 is Yes, the output unit 47 of the predictive maintenance device 400 presents the COP before and after the maintenance work to the customer (S33). The timing of this presentation may be when the SE visits the customer again. If a certain period of time has elapsed since the maintenance work, the output unit 47 of the predictive maintenance device 400 may send an email to the customer, and the COP before and after the maintenance work may be presented to the customer in that email. Alternatively, the output unit 47 may include a URL in the email, and when the customer's user terminal 5 connects to the URL, the COP before and after the maintenance work may be presented to the customer.

[0169] <Main Effects> The predictive maintenance device 400 of this disclosure determines whether the COP (Coefficient of Performance) has decreased when a failure is predicted in an air conditioner 40. If the COP has decreased, the priority of maintenance work on the air conditioner 40 is set higher than that of maintenance work on air conditioners 40 where the COP has not decreased, thereby improving the accuracy of the priority assigned to the air conditioner 40 in which a failure has been predicted. SEs can perform maintenance work on air conditioners 40 with the highest urgency first, reducing the number of cases that lead to failure. SEs can perform maintenance work according to priority, thus reducing the workload on the site.

[0170] <Other Application Examples> While the best mode for implementing this disclosure has been described above using examples, this disclosure is not limited to these examples, and various modifications and substitutions can be made without departing from the gist of this disclosure.

[0171] For example, in this disclosure, COP was used as the air conditioning capacity, but it is also acceptable to use the air conditioning capacity itself to determine whether or not the air conditioning capacity has decreased.

[0172] Furthermore, in this disclosure, the COP was calculated retrospectively when a failure was predicted, but the COP may be calculated continuously even before a failure is predicted.

[0173] Furthermore, the configuration examples shown in Figure 8 and other figures are divided according to their main functions to facilitate understanding of the processing performed by the predictive detection device 400. This disclosure is not limited by the way the processing units are divided or their names. The processing of the predictive detection device 400 can be further divided into many more processing units depending on the processing content. It can also be divided so that one processing unit includes even more processing.

[0174] Furthermore, the apparatus described in the examples represents only one of several computing environments for carrying out the embodiments disclosed herein. In one embodiment, the predictive detection device 400 includes multiple computing devices, such as a server cluster. The multiple computing devices are configured to communicate with each other via any type of communication link, including a network or shared memory, and perform the processing disclosed herein.

[0175] Each of the functions of this disclosure described above can be implemented not only by software processing through program execution, but also by one or more processing circuits. Here, "processing circuit" as used herein includes processors programmed to execute each function by software, such as processors implemented by electronic circuits, as well as devices such as ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), FPGAs (Field Programmable Gate Arrays), and conventional circuit modules designed to execute each of the functions described above.

[0176] <Reasons for the effect> The first aspect of this disclosure "detects signs of failure based on operating data, acquires the air conditioning capacity of the air conditioning system, determines the priority of maintenance work on the air conditioning system where signs of failure have been detected based on the air conditioning capacity, and outputs the priority along with the signs of failure." Therefore, the accuracy of the priority of maintenance work to be performed on an air conditioner where failure is predicted can be improved compared to determining the priority based on the part where signs of failure have been detected.

[0177] The second aspect of this disclosure "outputs the predicted time when a failure is likely to occur when signs of failure are detected," so that users can get an idea of ​​when a failure is likely to occur, making it easier for system engineers and others to determine maintenance work schedules.

[0178] A third aspect of this disclosure is that, "when signs of failure are detected, the probability of a failure occurring is output," which provides an indication of the likelihood of failure and makes it easier for SEs and others to decide which air conditioners should prioritize maintenance work on.

[0179] - The fourth aspect of this disclosure is that "when the difference between the expected air conditioning capacity and the measured air conditioning capacity is greater than a standard value, it is determined that the air conditioning capacity of the air conditioner has decreased based on predetermined judgment criteria," thus enabling accurate determination of whether the air conditioning capacity has decreased.

[0180] A fifth aspect of this disclosure is that "when signs of failure are detected and it is determined that the air conditioning capacity of the air conditioning system has decreased based on predetermined criteria, a message is output to the terminal indicating that maintenance work on the air conditioning system is to be proposed," so that the worker or user can understand that maintenance work should be performed because the air conditioning capacity of the air conditioning system has decreased.

[0181] A sixth aspect of this disclosure is that, "when the control unit receives notification that maintenance work has been completed for the air conditioning system, it calculates the air conditioning capacity of the air conditioning system after the maintenance work is completed." This allows the system to understand to what extent the air conditioning capacity of the air conditioning system has been restored by the maintenance work, and to demonstrate the effectiveness of the maintenance work to the customer.

[0182] - The seventh aspect of this disclosure is that "if the air conditioning capacity before maintenance work is determined to have decreased based on predetermined criteria, and the air conditioning capacity calculated after the completion of the maintenance work is determined not to have decreased based on predetermined criteria, then information regarding the air conditioning capacity of the air conditioning system is output," so that it is possible to appeal to the customer that the air conditioning capacity has returned to its original state due to the maintenance work.

[0183] The eighth aspect of this disclosure increases the priority of maintenance work on air conditioners in which signs of failure are detected and the air conditioning capacity of the air conditioner is judged to have decreased based on predetermined criteria, thereby increasing the priority of maintenance work on air conditioners that are at risk of failure.

[0184] The ninth aspect of this disclosure states that "when the rate of decline in air conditioning capacity is above a threshold, the priority of maintenance work on the air conditioning system in which signs of failure have been detected is increased compared to when the rate of decline in air conditioning capacity is below a threshold," thus allowing the priority of maintenance work on the air conditioner to be determined not only based on whether or not there is a decline in air conditioning capacity, but also on the rate of decline.

[0185] This application claims priority based on Japanese Patent Application No. 2024-201093, filed with the Japan Patent Office on November 18, 2024, and the entire contents of Japanese Patent Application No. 2024-201093 are incorporated herein by reference.

[0186] 40 Air conditioners 60 Server equipment 100 Predictive detection system 400 Predictive detection device

Claims

1. A predictive maintenance device for detecting signs of failure in an air conditioning system that is subject to maintenance work, comprising a control unit, wherein the control unit acquires operating data from the air conditioning system, detects signs of failure based on the operating data acquired from the air conditioning system, acquires the air conditioning capacity of the air conditioning system, determines the priority of maintenance work for the air conditioning system in which signs of failure have been detected based on the air conditioning capacity, and outputs the priority along with the signs of failure.

2. The predictive maintenance device according to claim 1, wherein the control unit outputs the time when a malfunction is predicted to occur when a malfunction is detected.

3. The predictive maintenance device according to claim 1, wherein the control unit outputs the probability of a failure occurring when a sign of failure is detected.

4. The control unit outputs an expected air conditioning capacity by inputting the operating data acquired from the air conditioning unit into a model that has learned the correspondence between the operating data acquired from the air conditioning unit and the air conditioning capacity; calculates the actual air conditioning capacity using the operating data acquired from the air conditioning unit; and determines that the air conditioning capacity of the air conditioning unit has decreased based on predetermined criteria if the difference between the expected air conditioning capacity and the actual air conditioning capacity is greater than a standard value, as described in claim 2 or 3.

5. The predictive maintenance device according to any one of claims 1 to 4, wherein the predictive maintenance device can communicate via a network with a terminal operated by a worker or a user of the air conditioning system, and the control unit outputs to the terminal a message proposing maintenance work for the air conditioning system when a sign of impending failure is detected and the air conditioning capacity of the air conditioning system is determined to be reduced based on predetermined criteria.

6. When a malfunction is detected and the control unit determines that the air conditioning capacity of the air conditioning unit has decreased based on predetermined criteria, the control unit receives notification that maintenance work has been completed for the air conditioning unit, the control unit calculates the air conditioning capacity of the air conditioning unit after the maintenance work has been completed, according to any one of claims 1 to 5.

7. The predictive maintenance device according to claim 6, wherein the control unit outputs information relating to the air conditioning capacity of the air conditioning system when it is determined that the air conditioning capacity before the maintenance work is performed has decreased based on predetermined criteria, and the air conditioning capacity calculated after the maintenance work is completed has not decreased based on predetermined criteria.

8. The pre-fault detection device according to any one of claims 1 to 7, wherein the control unit sets the priority of maintenance work for an air conditioner in which signs of failure have been detected and the air conditioning capacity of the air conditioner has been determined to have decreased based on predetermined criteria to be higher than the priority of maintenance work for an air conditioner in which signs of failure have been detected and the air conditioning capacity of the air conditioner has not been determined to have decreased based on predetermined criteria.

9. The predictive maintenance device according to any one of claims 1 to 8, wherein the control unit determines, based on predetermined criteria, that the air conditioning capacity of the air conditioning system is decreasing, determines the rate of decrease in air conditioning capacity, and if the rate of decrease in air conditioning capacity is greater than or equal to a threshold, it gives a higher priority to maintenance work on the air conditioning system in which a sign of failure has been detected than if the rate of decrease in air conditioning capacity is not greater than or equal to a threshold.

10. A predictive detection method for an air conditioning system that is subject to maintenance work, wherein the predictive detection device has a control unit, and the control unit performs: a process of acquiring operating data from the air conditioning system; a process of detecting a predictive failure based on the operating data acquired from the air conditioning system; a process of acquiring the air conditioning capacity of the air conditioning system; a process of determining the priority of maintenance work for the air conditioning system in which a predictive failure has been detected based on the air conditioning capacity; and a process of outputting the priority along with the predictive failure.

11. A program for causing a predictive maintenance device for detecting signs of failure in an air conditioning system that is subject to maintenance work to execute the following: a process for acquiring operating data from the air conditioning system; a process for detecting signs of failure based on the operating data acquired from the air conditioning system; a process for acquiring the air conditioning capacity of the air conditioning system; a process for determining the priority of maintenance work for the air conditioning system in which signs of failure have been detected based on the air conditioning capacity; and a process for outputting the priority along with the signs of failure.

Images