Fault detection method and server

Abstract

The invention relates to a fault detection method and a server. The method is applied to a server and comprises the steps that under the condition that fault declaration information for the air conditioner is received, equipment operation data of the air conditioner are obtained, and the fault declaration information comprises fault description data; based on the fault description data, the equipment operation data and a preset knowledge base, fault associated data and maintenance suggestion data corresponding to the air conditioner are detected; and generating a fault detection result of the air conditioner based on the fault associated data, the maintenance suggestion data and user portrait data corresponding to the air conditioner. By adopting the air conditioner, the fault solving efficiency of the air conditioner can be improved.

Description

Technical Field

[0001] This application relates to the field of air conditioner technology, and in particular to a fault detection method and server. Background Technology

[0002] With the rapid development and widespread application of air conditioner technology, the structure and function of air conditioners are becoming increasingly complex, and the types of equipment failures are becoming more diversified. These failures not only affect equipment performance but also lead to a continuous increase in the daily operation and maintenance pressure on air conditioner manufacturers.

[0003] In related technologies, when an air conditioner malfunctions, after-sales service personnel need to bring the appropriate tools to the user's home for inspection and repair based on the user's description of the malfunction.

[0004] However, it takes time for maintenance personnel to travel to and from the site. If a fault is misdiagnosed or the tools are not fully equipped, multiple trips to the site may be required. This not only makes it difficult to resolve the fault for a long time, but also delays the waiting time for other faults to be resolved, resulting in a decrease in the overall fault resolution efficiency. Summary of the Invention

[0005] This application provides a fault detection method and server to solve the problem of low efficiency in troubleshooting air conditioner faults.

[0006] In a first aspect, some embodiments provide a fault detection method applied to a server, the method comprising:

[0007] Upon receiving a fault report for an air conditioner, the system acquires the air conditioner's equipment operation data, including fault report data.

[0008] Based on fault description data, equipment operation data, and a pre-built knowledge base, detect the fault-related data and maintenance suggestion data corresponding to the air conditioner;

[0009] Based on fault correlation data, maintenance suggestion data, and user profile data corresponding to the air conditioner, fault detection results for the air conditioner are generated.

[0010] Technical Benefits: First, considering the inherent subjectivity of fault description data and its accuracy depending on the user's observation and expression abilities, while fault description data can intuitively and comprehensively reflect the abnormal performance of the air conditioner, equipment operation data itself is more objective and accurate. However, equipment operation data is ambiguous, as different faults may produce the same or similar equipment operation data. Therefore, by combining subjective fault description data with objective equipment operation data, key information can be cross-validated, overcoming the limitations of a single data source and significantly improving the accuracy and reliability of fault diagnosis. Second, by matching fault-related data and maintenance suggestion data from a pre-built knowledge base, a smaller amount of more relevant knowledge information can be filtered out and applied to fault detection. This not only effectively reduces the inference complexity of fault detection and improves fault detection efficiency but also enhances the accuracy of fault detection. Third, by combining fault detection with user profiles, more accurate fault detection results can be further inferred from the possible fault-related data and feasible maintenance suggestion data filtered from the knowledge base, making fault judgments that better match the user profile data and providing more suitable maintenance suggestions for the user. In this way, on the one hand, automated fault detection can reduce the time that maintenance personnel spend on fault detection and improve the efficiency of fault resolution; on the other hand, by improving the accuracy of automated fault detection, misjudgments can be reduced, thereby reducing the number of times maintenance personnel need to travel to the site and effectively improving the efficiency of fault resolution.

[0011] Secondly, some embodiments also provide a server, including: a storage module, a communication module, and a processor. The storage module is configured to store interface data and a delivery strategy; the interface data includes one or more combinations of a first template, a second template, and recommendation information; the communication module is configured to establish a communication connection with an air conditioner; the processor is configured to:

[0012] Upon receiving a fault report for an air conditioner, the system acquires the air conditioner's equipment operation data, including fault report data.

[0013] Based on fault description data, equipment operation data, and a pre-built knowledge base, detect the fault-related data and maintenance suggestion data corresponding to the air conditioner;

[0014] Based on fault correlation data, maintenance suggestion data, and user profile data corresponding to the air conditioner, fault detection results for the air conditioner are generated.

[0015] Technical Benefits: First, considering the inherent subjectivity of fault description data and its accuracy depending on the user's observation and expression abilities, while fault description data can intuitively and comprehensively reflect the abnormal performance of the air conditioner, equipment operation data itself is more objective and accurate. However, equipment operation data is ambiguous, as different faults may produce the same or similar equipment operation data. Therefore, by combining subjective fault description data with objective equipment operation data, key information can be cross-validated, overcoming the limitations of a single data source and significantly improving the accuracy and reliability of fault diagnosis. Second, by matching fault-related data and maintenance suggestion data from a pre-built knowledge base, a smaller amount of more relevant knowledge information can be filtered out and applied to fault detection. This not only effectively reduces the inference complexity of fault detection and improves fault detection efficiency but also enhances the accuracy of fault detection. Third, by combining fault detection with user profiles, more accurate fault detection results can be further inferred from the possible fault-related data and feasible maintenance suggestion data filtered from the knowledge base, making fault judgments that better match the user profile data and providing more suitable maintenance suggestions for the user. In this way, on the one hand, automated fault detection can reduce the time that maintenance personnel spend on fault detection and improve the efficiency of fault resolution; on the other hand, by improving the accuracy of automated fault detection, misjudgments can be reduced, thereby reducing the number of times maintenance personnel need to travel to the site and effectively improving the efficiency of fault resolution. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This application provides schematic diagrams of the architecture of an air conditioning system for some embodiments.

[0018] Figure 2 This is a schematic diagram of the external structure of an air conditioner provided in some embodiments of this application;

[0019] Figure 3 A flowchart illustrating the implementation of a fault detection method provided in some embodiments of this application;

[0020] Figure 4 A flowchart illustrating the knowledge base retrieval process provided for some embodiments of this application;

[0021] Figure 5 A flowchart illustrating the initial classification result fusion process provided in some embodiments of this application;

[0022] Figure 6 A flowchart illustrating the process of achieving target confidence detection provided in some embodiments of this application;

[0023] Figure 7 This is a flowchart illustrating the process of generating fault detection results provided in some embodiments of this application;

[0024] Figure 8 A flowchart illustrating the implementation of the knowledge base construction process provided in some embodiments of this application;

[0025] Figure 9 A flowchart illustrating the implementation of a fault detection method provided in other embodiments of this application;

[0026] Figure 10 This is a schematic diagram of the architecture of a fault detection server provided in some embodiments of this application. Detailed Implementation

[0027] The embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described below do not represent all embodiments consistent with this application. They are merely examples of systems and methods consistent with some aspects of this application as detailed in the claims.

[0028] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0029] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.

[0030] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.

[0031] The term "module" refers to any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0032] In the embodiments of this application, such as Figure 1 As shown, the air conditioning system 100 provided in this application includes a server 101 and an air conditioner 102. The air conditioner 102 and the server 101 communicate with each other via various communication methods. The air conditioner 102 can be connected to other networks via a local area network (LAN), a wireless local area network (WLAN), and other networks.

[0033] The air conditioner 102 can be a split-type wall-mounted air conditioner, a split-type floor-standing air conditioner, a ceiling-mounted air conditioner, a built-in air conditioner, a central air conditioner, or a multi-split air conditioner. This application embodiment does not impose any restrictions on this.

[0034] like Figure 2 As shown, the external structure of the air conditioner includes an indoor unit 201 and an outdoor unit 202. The indoor unit 201 is used to regulate the temperature and humidity of the indoor air. The outdoor unit 202 is connected to the indoor unit 201 through a connecting pipe. The indoor unit 201 is generally installed indoors, and the outdoor unit 202 is generally installed outdoors.

[0035] With the rapid development and widespread application of air conditioner technology, the structure and function of air conditioners are becoming increasingly complex, and the types of equipment failures are becoming more diversified. These failures not only affect equipment performance but also lead to a continuous increase in the daily operation and maintenance pressure on air conditioner manufacturers.

[0036] In related technologies, air conditioner maintenance mainly relies on user reports and regular inspections. When an air conditioner malfunctions, after-sales service personnel need to bring the appropriate tools to the user's home to inspect and repair it based on the fault description reported by the user.

[0037] However, for faults that can be quickly repaired through simple operations, users are more efficient at troubleshooting themselves than by having a technician come to their home. In hot or cold seasons, air conditioner malfunctions can severely impact users' quality of life and cause inconvenience. Higher troubleshooting efficiency and shorter downtime mean less negative impact on users' daily lives. For manufacturers, having repair manpower tied up in low-difficulty faults leads to delays in responding to high-difficulty faults that truly require on-site service, resulting in lower overall troubleshooting efficiency and reduced resource utilization.

[0038] For more complex faults, if the user's fault description is inaccurate or omits key information, it can easily lead to repair personnel misdiagnosing the fault type before heading to the site. This could result in repair personnel bringing the wrong tools or failing to equip themselves with all necessary spare parts. This would necessitate multiple trips to the site, not only delaying the resolution of the current fault but also causing delays in other pending faults, ultimately reducing overall fault-solving efficiency.

[0039] Based on this, embodiments of this application provide a fault detection method, which is applied to a server. For example... Figure 3 As shown, the fault detection method includes steps S10 to S30. Wherein:

[0040] Step S10: Upon receiving fault report information for the air conditioner, obtain the equipment operation data of the air conditioner, wherein the fault report information includes fault description data.

[0041] The fault report information refers to information submitted by the user related to an air conditioner malfunction, which may include at least one of the following: fault description data, fault occurrence time, etc. Fault report information can be submitted via application software, telephone, email, etc. If the air conditioner has a communication module, the fault report information can also be submitted directly through the air conditioner. Fault report information can be submitted directly by the user, or it can be submitted on behalf of after-sales personnel or repair personnel; this embodiment does not impose any restrictions on this. For example, when a user discovers a malfunction in the air conditioner, they can submit fault report information through a mobile application.

[0042] Fault description data refers to subjective descriptions by users of phenomena caused by air conditioner malfunctions. Fault description data can be in at least one form, such as text, voice, images, and video. For example, a user might first call customer service and describe the fault over the phone. With the user's permission, the customer service representative might record the fault description. Subsequently, under the representative's guidance, the user might send images and videos of the air conditioner in its faulty state via a mobile application or messaging app. In this case, the phone recording, the images, and the video sent by the user can all be considered as fault description data.

[0043] Fault description data can reflect the fault performance of an air conditioner relatively intuitively and comprehensively, but it has a certain degree of subjectivity, and its accuracy depends on the user's observation and expression abilities. By utilizing multimodal fault description data, the impact of user subjectivity, observation, and expression abilities on the fault description data can be effectively reduced, thereby improving the accuracy of the fault description data. However, the depth of fault description data is still insufficient; relying solely on fault description data cannot correspond to technical parameters and may overlook key details, resulting in lower accuracy of fault detection results.

[0044] Equipment operation data can refer to parameters recorded during the operation of the air conditioner, which may include at least one of the following: set temperature, indoor temperature, outdoor temperature, humidity, compressor operating frequency, fault self-test results of the air conditioner's built-in fault detection program, and equipment sound data.

[0045] Equipment operation data is collected and analyzed by the air conditioner. This data can be uploaded to the server in real time, or stored in the air conditioner's memory and uploaded to the server periodically.

[0046] In some feasible implementations, since the storage space of the air conditioner is relatively limited compared to that of the server, the device operation data uploaded to the server can be deleted from the air conditioner's storage.

[0047] While equipment operating data is inherently more objective and accurate, it is also ambiguous, as different types of faults can produce similar or identical operating data. Air conditioners, as complex mechatronic systems, have closely coupled subsystems. When a component malfunctions, its impact is transmitted through the system, ultimately manifesting as correlated changes in multiple operating parameters. For example, "high evaporator temperature and low compressor current" could be caused by insufficient refrigerant due to refrigerant leakage, insufficient expansion valve opening, or evaporator blockage; the repair solutions for these three are completely different. If the user describes the data as "cooling normally when first turned on, gradually worsening after one hour," combining these two factors strongly suggests expansion valve ice blockage, which initially works but gradually becomes blocked as moisture freezes.

[0048] Therefore, by combining subjective fault description data with objective equipment operation data, key information can be cross-validated, making up for the limitations of a single data source, thereby significantly improving the accuracy and reliability of fault diagnosis.

[0049] For example, if a user discovers an abnormality in the air conditioner while using it, or if a repairman discovers an abnormality during routine maintenance, they can report the fault through mobile application software, telephone, email, or other means. The fault report information can be uploaded to the server directly through the application software or by after-sales personnel.

[0050] Upon receiving a fault report, the server can first retrieve the corresponding device operation data from the storage of the air conditioner or server based on the fault time information contained in the fault report.

[0051] In some feasible implementations, the equipment operation data corresponding to the fault time information can refer to the equipment operation data within a certain period of time before the time indicated by the fault time information.

[0052] In some other feasible implementations, the equipment operation data corresponding to the fault time information may refer to the equipment operation data within a certain period of time from before the time indicated by the fault time information to after the time indicated by the fault time information.

[0053] In some feasible implementations, the fault time information in the fault report information can be provided by the user.

[0054] In other feasible implementations, since users typically stop using the air conditioner after it malfunctions, and even if they do continue to use it, the unresolved malfunction will still be reflected in the equipment's operating data. Therefore, if the user does not provide this information, the time of the malfunction report can be determined as the malfunction time.

[0055] Step S20: Based on fault description data, equipment operation data, and a pre-set knowledge base, detect the fault-related data and maintenance suggestion data corresponding to the air conditioner.

[0056] The pre-built knowledge base can refer to a structured data set used to store, organize, and retrieve professional knowledge, rules of experience, and solutions in the field of air conditioner malfunctions.

[0057] In some feasible implementations, fault detection results can be generated using a large language model, and the pre-built knowledge base can refer to a RAG (Retrieval-Augmented Generation) knowledge base. A RAG knowledge base is an intelligent knowledge base system with retrieval and generation capabilities. It can be used to dynamically acquire external knowledge and generate accurate responses. It can enhance the generation capabilities of the large language model by retrieving relevant data to ensure that the output results are both professional and verifiable.

[0058] Large Language Models (LLMs) are deep learning-based natural language processing models capable of understanding and generating human language. When combined with the RAG knowledge base, LLMs not only rely on the knowledge gained during training but also dynamically retrieve relevant information from the RAG knowledge base, thereby improving the accuracy, timeliness, and professionalism of their responses. This combination enables the model to provide more reliable and evidence-based outputs more effectively when facing domain-specific problems.

[0059] The training process of a large language model can include pre-training, fine-tuning, and incremental learning. During pre-training, the model learns language patterns from trillions of general-purpose texts, building basic language capabilities by predicting mask words or context. In fine-tuning, historical data from the air conditioner domain can be used to further train the model, adapting it to the specific task of air conditioner fault classification. After fine-tuning, the large language model can be deployed in real-world applications. During these applications, user feedback on the output results can be obtained to further adjust model parameters through incremental learning, optimizing the large language model's processing capabilities in complex real-world scenarios that are difficult to model in advance.

[0060] Fault-related data can refer to a set of data used to describe and characterize fault attributes and features, and may include at least one of the following: fault classification results, fault level, fault code, phenomenon description data, fault judgment index thresholds, and historical maintenance records. The phenomenon description data can be formulated and uploaded by professional maintenance personnel based on experience, or historical fault description data can be stored as phenomenon description data in a knowledge base. The fault judgment index threshold can refer to an index threshold formulated for one or more equipment operating data related to fault judgment, and can be determined based on experience or test results; this embodiment does not impose any limitations on this.

[0061] Maintenance recommendation data can refer to solutions for each type of fault and may include at least one of the following: operating procedures and replacement parts.

[0062] As an example, fault description data and equipment operation data can be used as indexes to retrieve fault-related data and maintenance suggestions for the air conditioner from a pre-built knowledge base.

[0063] As another example, the fault type can be determined first based on the fault description data and equipment operation data to identify the fault type of the air conditioner. Then, based on the fault type, the fault-related data and maintenance suggestion data corresponding to the air conditioner can be retrieved from the pre-set knowledge base.

[0064] As another example, fault association data includes fault classification results. In this case, fault type can also be determined based on equipment operation data to obtain a first initial classification result corresponding to the equipment operation data; simultaneously, fault description data is used as an index to retrieve from a pre-built knowledge base to determine a second initial classification result corresponding to the fault description data; then, the first and second initial classification results are merged to obtain the fault classification result for the air conditioner, and based on this fault classification result, the pre-built knowledge base is retrieved to obtain the fault association data and maintenance suggestion data corresponding to the air conditioner.

[0065] Step S30: Based on fault association data, maintenance suggestion data, and user profile data corresponding to the air conditioner, generate the fault detection result of the air conditioner.

[0066] User profile data can refer to a digital model that describes user characteristics, and may include at least one of the following: age, gender, technical skills, equipment model, and historical maintenance records.

[0067] For faults that can be quickly fixed with simple operations, user self-repair is more efficient than on-site repair by technicians. During hot or cold seasons, air conditioner malfunctions and downtime can severely impact users' quality of life and cause inconvenience. Higher fault-solving efficiency and shorter downtime minimize the negative impact on users' daily lives. For manufacturers, this also effectively improves resource utilization, allowing more repair personnel to be dedicated to resolving complex faults that truly require on-site service.

[0068] However, different users have different abilities to repair faults themselves. For users with professional technical backgrounds, strong learning and hands-on skills, they can quickly repair some faults that can be fixed by simple operations, which can effectively shorten the downtime of the air conditioner and thus effectively reduce the adverse effects of the fault on the user.

[0069] Therefore, by combining user profile data to provide final repair suggestions and generate personalized repair solutions, the accuracy and efficiency of repair solutions can be effectively improved. This not only enhances the user experience but also improves the utilization rate of repair resources. For example, suppose a user profile indicates that the user is "hands-on and skilled at repairs." For some easily repairable faults, the server can push self-service operation guides or operation videos to guide the user to try to repair the fault themselves first.

[0070] In some feasible implementations, user profile data can be obtained from associated terminals. For example, when a user uses the air conditioner's application software, they can log in to their own account, which can also log in to other applications or be associated with other applications. With the user's authorization, the air conditioner's application software can share data from other applications. If other applications have already generated the user's user profile data, it can be directly obtained and applied.

[0071] In other feasible implementations, user profile data can be generated based on at least one of user behavior data, historical maintenance records, and user characteristic data. User behavior data can include at least one of user characteristic data such as usage habits, operation records, voiceprint recognition, and infrared sensor recognition. For example, if a user previously reported a malfunction to the air conditioner or other home appliance, generating a historical maintenance record, and this record shows the user repairing the malfunction themselves with the guidance of a repairman over the phone, or if a repairman discovered the user's self-repair capabilities during a site visit and recorded this in the historical maintenance record, then it can be determined that the user is capable of repairing some easily repairable malfunctions independently.

[0072] Fault detection results can refer to the diagnostic conclusions and response plans ultimately generated from fault report information. Diagnostic conclusions can include at least one of the following: fault classification results, confidence levels of the fault classification results, and diagnostic analysis. Diagnostic analysis refers to the reasoning process that derives the final diagnostic conclusion based on fault description data and equipment operating data. Through diagnostic analysis, the transparency, reliability, and usability of the fault handling system can be significantly improved. When the diagnostic conclusion is inaccurate, the diagnostic analysis logic can be traced back to correct the knowledge base rules and optimize the model. It can also help users and maintenance personnel correct errors in a timely manner.

[0073] As an example, fault correlation data, maintenance suggestion data, and user profile data corresponding to the air conditioner can be input into a pre-trained large language model to generate fault detection results for the air conditioner.

[0074] As another example, fault characteristics and solutions can be matched using predefined logical rules. If there are multiple matching solutions, one can be selected as the output based on the user profile. Then, the fault characteristics and solutions are filled into the predefined fault detection result template to generate the fault detection result.

[0075] In this embodiment, firstly, considering that fault description data has a certain degree of subjectivity and its accuracy depends on the user's observation and expression abilities, although it can intuitively and comprehensively reflect the abnormal performance of the air conditioner, and equipment operation data itself is more objective and accurate, it is also ambiguous, as different faults may produce the same or similar equipment operation data. Therefore, by combining subjective fault description data with objective equipment operation data, key information can be cross-validated, overcoming the limitations of a single data source, thereby significantly improving the accuracy and reliability of fault diagnosis. Secondly, by matching fault-related data and maintenance suggestion data from a pre-built knowledge base, a small amount of more relevant knowledge information can be filtered out and applied to fault detection. This not only effectively reduces the reasoning complexity of fault detection and improves fault detection efficiency, but also enhances the accuracy of fault detection. Thirdly, by combining user profiles for fault detection, more accurate fault detection results can be further inferred from the possible fault-related data and feasible maintenance suggestion data filtered from the knowledge base, making fault judgments that better match the user profile data and providing more suitable maintenance suggestions for the user. In this way, on the one hand, automated fault detection can reduce the time that maintenance personnel spend on fault detection and improve the efficiency of fault resolution; on the other hand, by improving the accuracy of automated fault detection, misjudgments can be reduced, thereby reducing the number of times maintenance personnel need to travel to the site and effectively improving the efficiency of fault resolution.

[0076] In some embodiments, such as Figure 4As shown, the fault association data includes fault classification results; based on fault description data, equipment operation data, and a pre-built knowledge base, the fault association data and maintenance suggestion data corresponding to the air conditioner are detected, including steps S21 to S23. Among them:

[0077] Step S21: Based on the fault description data, match the first initial classification result corresponding to the air conditioner from the preset knowledge base, and perform fault detection on the air conditioner based on the equipment operation data to obtain the second initial classification result.

[0078] It's important to note that knowledge base matching is typically based on semantic similarity. However, most device operational data consists of numerical parameters, whose determination usually relies on physical causality. Semantic matching, on the other hand, is essentially statistical similarity. Forcing physical parameter thresholds and other rules into semantics is akin to using statistical methods to solve a physics problem, making it difficult to guarantee the same level of accuracy as logical judgment. This can lead to a significant increase in the false positive rate. Furthermore, for fault description data, logical judgment often struggles to adapt to the diversity of user expressions.

[0079] The first initial classification result can refer to the fault type initially determined based on fault description data.

[0080] The second initial classification result can refer to the fault type derived from preliminary analysis of equipment operation data.

[0081] The fault classification result can refer to the fault type determined by comprehensively considering fault description data and equipment operation data.

[0082] For example, after obtaining fault report information, the server can first extract fault description data from the fault report information, and then perform semantic similarity matching between the fault description data and a pre-set knowledge base, and determine at least one fault type whose semantic similarity meets the preset similarity conditions as the first initial classification result.

[0083] After obtaining the device operation data, the server can classify the device operation data into at least one fault type through logical reasoning or through a trained machine learning model, and determine the fault type determined by logical classification as the second initial classification result.

[0084] In some feasible implementations, the preset similarity condition can refer to the highest similarity; that is, the fault type corresponding to the highest semantic similarity is determined as the first initial classification result.

[0085] In other feasible implementations, the preset similarity condition may refer to a value higher than a preset similarity threshold; that is, the fault type corresponding to the semantic similarity value higher than the preset similarity threshold is determined as the first initial classification result.

[0086] In some feasible implementations, the logical reasoning method may include classifying according to a unified logical classification rule; or detecting the matching degree between the equipment operation data and each fault type according to the fault classification logical rule corresponding to each fault type; or classifying through a preset logical reasoning model; etc., this embodiment does not limit this.

[0087] Step S22: Combine the first initial classification result and the second initial classification result to obtain the fault classification result.

[0088] For example, after determining the first initial classification result and the second initial classification result, the first initial classification result and the second initial classification result can be logically integrated to obtain the fault classification result.

[0089] In some feasible implementations, the logical integration of the first initial classification result and the second initial classification result may include: if at least some of the fault types in the first initial classification result and the second initial classification result are the same, the same fault type may be identified as the fault classification result.

[0090] In some feasible implementations, the logical integration of the first initial classification result and the second initial classification result may include: when the first initial classification result and the second initial classification result are inconsistent, and there is no confidence information in the first initial classification result and the second initial classification result, since the fault description data is highly subjective and key information is easily missed, the second initial classification result can be determined as the fault classification result, and the classification result of the objective equipment operation data shall prevail.

[0091] In this scenario, the initial classification result can be used as a reference fault type. When displaying the fault detection results, the reference fault type can be shown simultaneously, allowing maintenance personnel, after-sales staff, or users to make a manual judgment based on the fault detection results and the reference fault type. Maintenance personnel or after-sales staff can guide users to supplement fault description data or provide more precise fault description data based on the fault detection results and the reference fault type. After receiving new fault description data, the system can return to the steps of detecting the corresponding fault-related data and maintenance suggestion data for the air conditioner based on the fault description data, equipment operating data, and a pre-built knowledge base. The fault detection results are then regenerated based on the equipment operating data and the updated fault description data.

[0092] In some feasible implementations, the logical integration of the first and second initial classification results can include: when the first and second initial classification results are inconsistent, but confidence information exists in both, fault classification results can be selected from the first and second initial classification results based on the confidence information. For example, the fault type with the highest confidence in the first and second initial classification results can be determined as the fault classification result. Alternatively, the highest confidence level in the first initial classification result can be compared with the highest confidence level in the second initial classification result. If the highest confidence level in the second initial classification result is lower than the highest confidence level in the first initial classification result, and the highest confidence level in the second initial classification result is lower than a preset confidence threshold, then the fault type corresponding to the highest confidence level in the first initial classification result is determined as the fault classification result; otherwise, the fault type corresponding to the highest confidence level in the second initial classification result is determined as the fault classification result.

[0093] In some feasible implementations, the logical integration of the first initial classification result and the second initial classification result may include: if at least some fault types are the same in the first initial classification result and the second initial classification result, and confidence information exists in both the first and second initial classification results, the confidence levels corresponding to each fault type can be fused to obtain a target confidence level for each fault type, and then the fault type corresponding to the highest target confidence level can be determined as the fault classification result. The confidence level fusion method may include summation, averaging, weighted summation, or weighted averaging, etc., and this embodiment does not impose any limitations on this.

[0094] Step S23: Query the maintenance suggestion data corresponding to the fault classification results from the preset knowledge base.

[0095] For example, after determining the fault classification result, the maintenance suggestion data corresponding to the fault classification result can be accurately retrieved from the preset knowledge base.

[0096] In this embodiment, a hybrid classification strategy is employed to semantically match fault description data with a pre-built knowledge base. This leverages the synonym generalization and fuzzy matching capabilities of semantic matching to accurately match fault description data with synonymous polymorphism and ambiguous information, thereby improving the accuracy of fault classification based on fault description data. For equipment operation data, logical rules can be used to convert numerical equipment operation data into classification results, further improving the accuracy of fault classification based on equipment operation data. By integrating both subjective fault description data and objective equipment operation data, the limitations of a single data source are overcome, further enhancing the accuracy of fault classification. Consequently, based on the semantic fault classification results, corresponding maintenance suggestion data can be accurately matched from the knowledge base.

[0097] In some embodiments, fault detection of the air conditioner is performed based on equipment operating data to obtain a second initial classification result, including:

[0098] Based on the pre-defined fault classification logic rules corresponding to various fault types, the fault type that matches the equipment operation data is detected, and a second initial classification result is obtained.

[0099] It should be noted that air conditioners generate a wide variety of operational data, with large volumes and numerous fault types. A single logical judgment rule is insufficient to accurately diagnose each fault type. Machine learning models not only require significant computing power but also exhibit poor scalability. As technology advances, the types of operational data and fault types may further increase. This increase in data types and fault types typically necessitates retraining the machine learning model with even greater resources.

[0100] Among them, the fault classification logic rules can refer to a set of predefined logical judgment conditions for a specific fault type, which may include constraints such as thresholds and timing relationships of equipment operation data.

[0101] In some feasible implementations, the fault classification logic rule may include at least one of triggering conditions, fault type labels, and confidence weights. The triggering condition may be a logical combination of device parameters. The fault type label may be used as the output result during rule matching. The confidence weight may refer to the rule's credibility score, which can be determined based on experience or test results, and this embodiment does not impose any limitations on it.

[0102] In some feasible implementations, an expert mini-model can be used to detect the fault type matched by the equipment operating data to obtain a second initial classification result. The expert mini-model includes multiple sub-models, with different fault types corresponding to different sub-models. Each sub-model is used to detect whether the equipment operating data matches a specific fault type. For each sub-model, it performs logical analysis and judgment on the equipment operating data according to the fault classification logic rules corresponding to that specific fault type, thereby detecting whether this equipment operating data can characterize the air conditioner for a fault corresponding to that specific fault type.

[0103] Expert mini-models can use labeled historical equipment operation data as training samples for model training, thereby optimizing model parameters such as decision thresholds in the fault classification logic rules and improving the accuracy of fault classification. Specifically, fault classification logic formulas corresponding to each fault type can be formulated based on experience, and the model parameters in the fault classification logic formulas can be initialized to obtain the expert mini-model to be trained. Then, historical equipment operation data is input into the expert mini-model to be trained to obtain fault classification training results. Next, the fault classification training results are compared with the classification labels pre-labeled for the historical equipment operation data, and the gradient value is calculated based on the difference between the two. Then, based on the gradient value, the expert mini-model is iteratively optimized through backpropagation.

[0104] Compared to large models, expert small models are more lightweight, more efficient, and offer greater specialization and interpretability for specific fault types, enabling faster and more accurate identification and response to corresponding fault modes. Since equipment operating data is primarily numerical, with clear physical meaning and patterns of change, expert small models rely more on interpretable logical rules and actual data characteristics for fault detection. This reduces dependence on complex semantic understanding and contextual reasoning, making them less prone to "illusions" compared to large models, resulting in more stable and reliable output.

[0105] For example, after obtaining the equipment operation data, the equipment operation data can be matched with the fault classification logic rule corresponding to each fault type. If a target fault classification logic rule that matches the equipment operation data is detected, the fault type corresponding to the target fault classification logic rule can be determined as the second initial classification result.

[0106] In some feasible implementations, there may be multiple target fault classification logic rules that match the equipment operation data. In this case, all target fault classification logic rules can be determined as the second initial classification result; or, a candidate list can be generated based on the fault type corresponding to each target fault classification logic rule, and the fault type with the highest confidence weight can be selected from the candidate list as the second initial classification result based on the confidence weight.

[0107] In this embodiment, by setting corresponding fault classification logic rules for each fault type, the fault classification logic rules for different fault types are independent of each other and do not affect each other. This ensures accurate determination of each fault type. When a new fault type is added, only rule entries need to be added without modifying the core algorithm, which has high scalability.

[0108] In some embodiments, the first initial classification result and the second initial classification result are fused to obtain the fault classification result, including:

[0109] If the first initial classification result is inconsistent with the second initial classification result, and there is no confidence information in either the first or second initial classification result, the second initial classification result shall be determined as the fault classification result.

[0110] It's important to note that the initial classification results typically undergo a preliminary screening. For example, the first initial classification result usually only outputs fault types with semantic relevance exceeding a certain threshold. Inaccurate fault descriptions or missing information by users can easily lead to a significant decrease in the accuracy of the first initial classification result. For instance, suppose there are two sounds, A and B, that are very similar in description. Sound A is more common and corresponds to fault A, while sound B is less common and corresponds to fault B. When a user describes the sound of fault B, they might not immediately recall sound B and simply say it's similar to sound A. In this case, semantic similarity matching will be more likely to match fault A, leading to a misclassification.

[0111] Confidence information can refer to a quantitative assessment of the credibility of the current fault classification result.

[0112] For example, when the first initial classification result is inconsistent with the second initial classification result, and there is no confidence information in the first initial classification result and the second initial classification result, there is uncertainty in both the fault description data and the semantic matching process. However, the equipment operation data itself is relatively accurate. Therefore, the second initial classification result determined based on the equipment operation data can be determined as the final fault classification result.

[0113] In some feasible implementations, the initial classification result can also be used as a reference fault type. When displaying the fault detection result, the reference fault type can be displayed simultaneously, allowing maintenance personnel, after-sales personnel, or users to make a manual judgment based on the fault detection result and the reference fault type. Maintenance personnel or after-sales personnel can guide users to supplement fault description data or provide more accurate fault description data based on the fault detection result and the reference fault type. After receiving new fault description data, the process can return to the steps of detecting the corresponding fault-related data and maintenance suggestion data of the air conditioner based on the fault description data, equipment operation data, and a pre-built knowledge base. The fault detection result is then regenerated based on the equipment operation data and the updated fault description data.

[0114] In this embodiment, when the first initial classification result and the second initial classification result are inconsistent, the judgment result based on objective equipment operation data is more accurate. Therefore, the second initial classification result based on objective equipment operation data can be determined as the fault classification result and displayed to the user. Furthermore, to avoid the second initial classification result being a misclassification, the first initial classification result can also be displayed simultaneously as a reference, but its corresponding maintenance suggestions are not retrieved or displayed at this time. In this way, if the second initial classification result does not misclassify, the resource consumption required to retrieve maintenance suggestion data can be effectively reduced; and if the second initial classification result does misclassify, maintenance personnel can refer to the first initial classification result to promptly identify and address the problem, reducing invalid maintenance operations based on misclassification results.

[0115] In some embodiments, such as Figure 5 As shown, the first initial classification result includes first confidence information corresponding to each of multiple first initial types; the second initial classification result includes second confidence information corresponding to each of multiple second initial types; each first initial type and each second initial type are at least partially consistent; fusing the first initial classification result and the second initial classification result to obtain the fault classification result includes steps S221 to S222. Wherein:

[0116] Step S221: Based on the first confidence information and the second confidence information, detect the target confidence corresponding to each first initial type and each second initial type.

[0117] It should be noted that the first initial classification result is determined based on fault description data, while the second initial classification result is determined based on equipment operation data. However, fault description data has a certain degree of subjectivity, and its accuracy depends on the user's observation and expression abilities, while equipment operation data is ambiguous; different types of faults may produce the same or similar equipment operation data. Therefore, the fault classification result selected from either the first or second initial classification result has relatively low accuracy.

[0118] The first confidence information can refer to the quantitative assessment of the credibility of the first initial classification result.

[0119] In some feasible implementations, the first initial classification result is determined by semantic similarity matching, that is, by performing semantic similarity matching between the fault description data and the phenomenon description data corresponding to each fault type. Therefore, the confidence information in the first initial classification result can be represented by semantic similarity; the higher the semantic similarity between the fault description data and the phenomenon description data corresponding to the fault type, the higher the confidence of that fault type.

[0120] The second confidence information can refer to a quantitative assessment of the credibility of the second initial classification result.

[0121] In some feasible implementations, the second initial classification result is determined through fault classification logic rules; that is, the equipment operating data is matched with the fault classification logic rules corresponding to each fault type. In this case, the equipment operating data typically only has two possibilities: a match or a non-match. However, the fault classification logic rules corresponding to each fault type may include multiple sub-rules. Each sub-rule can determine its corresponding confidence level based on experience or test results. The fused confidence level corresponding to the equipment operating data can be calculated based on the confidence levels corresponding to the sub-rules that the equipment operating data satisfies. The confidence level information in the second initial classification result can be represented by the fused confidence level. For example, the fault classification logic rules for fault type A include sub-rules r1, r2, and r3. The confidence level of r1 is 0.9, the confidence level of r2 is 0.8, and the confidence level of r3 is 0.5. Satisfying any two of these sub-rules is considered to belong to fault type A. The fault classification logic rules for fault type B include sub-rules r4 and r5. The confidence level of r4 is 0.9, and the confidence level of r5 is 0.7. Both sub-rules must be satisfied to be considered to belong to fault type B. If the equipment operation data satisfies r2, r3, r4, and r5, the fusion confidence level of fault type A can be 0.8 + 0.5 = 1.3, and the fusion confidence level of fault type B can be 0.9 + 0.7 = 1.6. Therefore, the confidence level information in the second initial classification result can include the fusion confidence level of fault type A (1.3) and the fusion confidence level of fault type B (1.6).

[0122] The initial type can refer to the fault type determined based on equipment operating data or fault description data. Initial types include a first initial type and a second initial type. The first initial type can refer to the fault type determined in the first initial classification result. The second initial type can refer to the fault type determined in the second initial classification result.

[0123] For example, after obtaining the first initial classification result, multiple first initial types are determined from the first initial classification result, and the first confidence information corresponding to each first initial type is extracted from the first initial classification result; after obtaining the second initial classification result, multiple second initial types are determined from the second initial classification result, and the second confidence information corresponding to each second initial type is extracted from the second initial classification result. Furthermore, for each initial type in the first and second initial types, its corresponding confidence information is fused to obtain the target confidence level corresponding to that initial type. For example, assuming the first initial type includes initial type A, and its corresponding first confidence level is T1, and the second initial type also includes initial type A, and its corresponding second confidence level is T2, T1 and T2 can be fused to obtain T3, and T3 can be used as the target confidence level for initial type A.

[0124] The confidence fusion method may include at least one of summation, averaging, weighted summation, weighted averaging, etc., and this embodiment does not limit it.

[0125] In some feasible implementations, if an initial type belongs to only one of the first initial classification result and the second initial classification result, the confidence information of classifying the initial type using another classification method can be further obtained, or the confidence of classifying the initial type using another classification method can be determined to a preset minimum value, such as 0.

[0126] Step S222: Determine the initial type corresponding to the highest target confidence as the fault classification result.

[0127] For example, after calculating the target confidence level for each initial type, the numerical values ​​of these target confidence levels are compared, and the initial type corresponding to the highest target confidence level is determined as the fault classification result.

[0128] In this embodiment, the target confidence score is generated by fusing confidence scores to make up for the limitations of a single data source. The complementary information from the two sources can maintain a relatively reliable output even when the data is abnormal or the description is ambiguous, thereby significantly improving the accuracy and reliability of fault diagnosis.

[0129] In some embodiments, such as Figure 6 As shown, based on the first confidence information and the second confidence information, the target confidence corresponding to each first initial type and each second initial type is detected, including steps S2211 to S2213. Wherein:

[0130] Step S2211: Determine the target initial type from each of the first initial types and each of the second initial types.

[0131] It should be noted that the first initial classification result is determined based on fault description data through semantic similarity matching, while the second initial classification result is determined based on equipment operation data through fault classification logic rules. Fault description data is highly subjective and easily overlooks key information, while equipment operation data is objective data with relatively high accuracy. Both fault description data and the semantic matching process have high uncertainty, while equipment operation data itself is relatively accurate. Therefore, it can be seen that the accuracy of the first and second initial classification results is not entirely the same. Although the initial classification result with lower accuracy is instructive for determining the final fault type, directly merging the two can easily lead to misjudgment due to the lower accuracy of the initial classification result.

[0132] For example, after obtaining the first initial classification result and the second initial classification result, multiple first initial types can be determined from the first initial classification result, and multiple second initial types can be determined from the second initial classification result; then, one of the first initial type and the second initial type is determined as the target initial type according to a preset order or randomly. It is understood that for the same first initial type and second initial type, only one of them can be used as the target initial type, and its corresponding target confidence level can be determined. Alternatively, the target confidence level can be calculated twice. This embodiment does not limit this.

[0133] It is understood that for each of the first initial type and the second initial type, steps S22111 to S2213 can be performed in parallel or sequentially to determine the target confidence level corresponding to each of the first initial type and the second initial type. This embodiment does not limit this.

[0134] Step S2212: Query the first confidence level corresponding to the initial type of the target from each first confidence level information, and query the second confidence level corresponding to the initial type of the target from each second confidence level information.

[0135] For example, after determining the initial type of the target, the first confidence level corresponding to the initial type of the target can be queried from the first confidence level information, and the second confidence level corresponding to the initial type of the target can be queried from the second confidence level information.

[0136] In some feasible implementations, if the first confidence level corresponding to the target initial type cannot be found from the first confidence level information, it may be because the first confidence level is low, resulting in the target initial type not being selected as the first initial type. In this case, the first confidence level may be stored in the memory corresponding to the first initial classification result, so the first confidence level corresponding to the target initial type can be obtained from the memory corresponding to the first initial classification result; or the first confidence level can be determined to a preset minimum value, such as 0.

[0137] In some feasible implementations, if the second confidence level corresponding to the target initial type cannot be found from the second confidence level information, it may be because the second confidence level is low, resulting in the target initial type not being selected as the second initial type. In this case, the second confidence level may be stored in the memory corresponding to the second initial classification result, so the second confidence level corresponding to the target initial type can be obtained from the memory corresponding to the second initial classification result; or the second confidence level can be determined to a preset minimum value, such as 0.

[0138] Step S2213: Based on the first weight corresponding to the first initial classification result and the second weight corresponding to the second initial classification result, the first confidence and the second confidence are weighted and fused to obtain the target confidence corresponding to the target initial type.

[0139] The first weight characterizes the importance of the first initial classification result to the fault classification. The second weight characterizes the importance of the second initial classification result to the fault classification.

[0140] In some feasible implementations, the allocation of the first weight and the second weight can be determined based on the classification accuracy of the first initial classification result and the second initial classification result. The classification accuracy of the first initial classification result and the second initial classification result can be determined based on experience, test results or historical data, etc., and this embodiment does not limit this.

[0141] For example, after determining the first confidence level and the second confidence level, the first weight corresponding to the first initial classification result is assigned to the first confidence level, and the second weight corresponding to the second initial classification result is assigned to the second confidence level. Then, the first weight is multiplied by the first confidence level, and the second weight is multiplied by the second confidence level. The two multiplications are then summed or averaged, and the result after fusion is determined as the target confidence level corresponding to the initial type of the target.

[0142] In this embodiment, by setting weights, it is possible to achieve complementary information from two sources, maintaining reliable output even when data is abnormal or descriptions are ambiguous; it is also possible to effectively reduce the probability of misjudgment caused by low-accuracy initial classification results, thereby improving the accuracy of fault detection.

[0143] In some embodiments, such as Figure 7 As shown, the fault association data includes fault classification results and initial fault grading results; based on the fault association data, maintenance suggestion data, and user profile data corresponding to the air conditioner, the fault detection results of the air conditioner are generated, including steps S31 to S32. Wherein:

[0144] Step S31: If the initial fault classification result contains multiple alternative fault classification results, determine the target fault classification result from each alternative fault classification result based on the user profile data corresponding to the air conditioner.

[0145] It should be noted that different users have varying abilities to repair malfunctions themselves. Users with professional technical backgrounds, strong learning and hands-on skills can quickly fix some simple problems themselves. In such cases, if repair personnel are still uniformly assigned to provide on-site repairs, it will only prolong the downtime of the air conditioner due to malfunctions. Users will be unable to use their air conditioners normally for a longer period. In hot or cold seasons, air conditioner malfunctions and downtime will seriously affect users' quality of life and cause inconvenience.

[0146] The initial fault classification result can refer to the fault classification result corresponding to each fault type retrieved from a pre-built knowledge base. Fault classification can be based on whether the user can repair it themselves. That is, the initial fault classification result can include a first fault level and a second fault level. The first fault level can refer to the fault level that the user can repair themselves, and the second fault level can refer to the fault level that requires on-site repair by a maintenance personnel.

[0147] In some feasible implementations, for faults that can be quickly repaired through simple operations, the initial fault classification result may include two alternative fault classification results: a first fault level and a second fault level. This allows for further classification based on user profile data.

[0148] In some feasible implementations, for faults that users cannot repair themselves, the corresponding initial fault classification result may only include the second fault level as an alternative fault classification result.

[0149] For example, after obtaining the initial fault classification result, it can be first determined whether the initial fault classification result contains multiple candidate fault classification results. If the initial fault classification result contains only one candidate fault classification result, then that candidate fault classification result can be determined as the target fault classification result. If the initial fault classification result contains multiple candidate fault classification results, then the corresponding user profile data can be queried based on the fault report information, and then, based on the preset mapping relationship between user profiles and fault classifications, the target fault classification result corresponding to that user profile data can be determined from the candidate fault classification results.

[0150] Step S32: Based on the fault classification results, target fault grading results, and maintenance suggestion data, generate the fault detection results of the air conditioner.

[0151] For example, the fault classification results, target fault classification results, and maintenance suggestion data can be filled into a predefined fault detection result template to generate fault detection results; or, the fault classification results, target fault classification results, and maintenance suggestion data can be input into a pre-trained large language model to generate fault detection results for the air conditioner.

[0152] In this embodiment, multiple alternative fault classifications are set for certain fault types, and then the most suitable maintenance suggestion data is matched to the user based on user profile data. In this way, users with professional technical backgrounds and strong learning and hands-on abilities can repair some faults that can be quickly fixed by simple operations by repairing the corresponding maintenance suggestion data themselves, thereby reducing the time these users spend waiting for maintenance personnel to come to their homes, shortening the downtime of the air conditioner due to faults, and improving the user experience.

[0153] In some embodiments, such as Figure 8 As shown, before detecting the fault-related data and maintenance suggestion data corresponding to the air conditioner based on fault description data, equipment operation data, and a pre-set knowledge base, the fault detection method further includes steps 802 to 806. Wherein:

[0154] Step 802: Obtain at least one of the following in the field of air conditioners: fault code corresponding information, fault judgment mode data, historical repair work order data, fault repair measures information, and fault component replacement information.

[0155] Among them, the fault code information can refer to the mapping relationship table between air conditioner error codes and fault types.

[0156] Fault diagnosis mode data can refer to a set of logical rules for diagnosing faults.

[0157] Historical maintenance work order data can refer to maintenance records that include fault symptoms, testing process, and final solution.

[0158] Fault repair information can refer to standard operating procedures for specific faults, and can be in the form of text, images, videos, etc.

[0159] Faulty component replacement information can refer to information about components associated with a specific fault.

[0160] In some feasible implementations, the faulty component replacement information can be a maintenance fault tree, which can include at least one of the following: fault classification, fault type, corresponding maintenance measures, and component recommendations. The correspondence between fault classification, fault type, corresponding maintenance measures, and component recommendations can be established in the form of tables or relationship diagrams.

[0161] For example, before actually performing fault detection, a knowledge base in the field of air conditioning can be built first. Specifically, at least one of the following can be obtained: fault code information corresponding to the air conditioner, fault judgment mode data, historical repair work order data, fault repair measures information, and faulty component replacement information.

[0162] In some feasible implementations, these data can also be standardized, such as standardizing the units and time formats of the data, and removing duplicate records and invalid data.

[0163] Step 804: Convert the fault code information, fault judgment mode data, historical repair work order data, fault repair measures information, and fault component replacement information into tagged text data.

[0164] Tagged text data can refer to text format with structured tags, which facilitates unified parsing.

[0165] For example, according to the preset marked text conversion rules, the fault code corresponding information, fault judgment mode data, historical maintenance work order data, fault maintenance measures information and faulty component replacement information are converted into marked text data in marked text format.

[0166] Step 806: Construct a knowledge base for the air conditioner domain based on tagged text data.

[0167] For example, after the marked text format is converted, a knowledge base for the air conditioner domain is constructed based on the marked text data in the obtained marked text format.

[0168] In this embodiment, by using structured storage of multi-source data such as fault code correspondence information, fault judgment mode data, historical maintenance work order data, fault repair measures information, and faulty component replacement information, accurate matching of fault characteristics can be achieved. Simultaneously, by constructing a knowledge base in the air conditioning field, more accurate and effective data support can be provided for fault detection, thereby reducing inference illusions.

[0169] In some embodiments, the fault detection results are generated by a large language model; after generating the fault detection results for the air conditioner based on fault association data, maintenance suggestion data, and user profile data corresponding to the air conditioner, the fault detection method further includes:

[0170] Upon receiving maintenance feedback data for air conditioners, the large language model is updated based on the maintenance feedback data.

[0171] Among them, maintenance feedback data can refer to the fault handling results record submitted by users, after-sales personnel or maintenance personnel after the air conditioner fault repair is completed. It can include information such as the actual fault type, repair measures, and replaced parts.

[0172] Large language models can refer to natural language processing models used for fault diagnosis, which can understand user descriptions and output maintenance suggestions.

[0173] For example, after generating the fault detection results for the air conditioner, these results can be displayed to the user. If the problem can be repaired independently, the fault detection results can include repair suggestions, such as a repair video. Within a certain period after sending the repair video, a repair feedback page can be sent to the user to obtain repair feedback data for cases where the user did not respond through the repair feedback process. If on-site repair is required, a repairman can be automatically assigned to the site. At the end of the repairman's on-site repair work order, a repair feedback page can be sent to at least one of the user and the repairman to obtain repair feedback data for cases where at least one of the user and the repairman did not respond through the repair feedback process.

[0174] After obtaining maintenance feedback data, a certain amount of maintenance feedback data can be collected first, and then the collected maintenance feedback data can be cleaned and labeled. Then, the cleaned and labeled maintenance feedback data can be used to incrementally train the large language model.

[0175] In this embodiment, incremental training based on maintenance feedback data can further improve the accuracy of fault detection, reduce the recurrence of previously observed misjudgment patterns, reduce repeated on-site visits due to model errors, and improve fault detection efficiency.

[0176] In some feasible implementations, such as Figure 9 As shown, the fault detection method may include the following steps:

[0177] Data collection. This includes collecting fault description data uploaded by users, equipment operation data uploaded by the air conditioner, and user profile data. Equipment operation data can be multimodal data collected in real time through the air conditioner's sensors and IoT modules, such as set temperature, indoor temperature, outdoor temperature, humidity, compressor operating frequency, fault occurrences, and device noise.

[0178] Data preprocessing. The collected fault description data is cleaned, standardized, and feature extracted to remove noisy data and outliers. Different types of data are standardized into a unified format and used as input to the RAG knowledge base in vectorized form.

[0179] Expert mini-model fault classification. The collected equipment operation data is input into the expert mini-model. Based on the unique equipment operation time series structured data of the air conditioner, the expert experience mini-model judges and outputs the fault type. The output fault type is then input into the RAG knowledge base.

[0180] The expert mini-model can first determine the fault type based on the fault codes and fault type correspondence table reported by the equipment. If the fault does not belong to the fault code table, it can then make a judgment based on the fault judgment mode and the returned equipment operating status parameter values. Based on the judgment logic in the expert mini-model, it can make a supplementary judgment on the fault type and finally output the fault type to the RAG knowledge base.

[0181] Matching with the RAG knowledge base. The fault description data input into the RAG knowledge base is semantically matched with the RAG knowledge base to determine the corresponding fault type, fault level, and maintenance recommendations. Simultaneously, expert mini-models are input into the fault types in the RAG knowledge base to determine the corresponding fault level and maintenance recommendations.

[0182] The fault classification can be divided into L1 level faults that users can repair themselves, and L2 level faults that require professional personnel to come to the site for repair, thus achieving fine-grained fault classification.

[0183] The construction of the RAG knowledge base may include: organizing historical maintenance records and maintenance fault trees into markdown (marked text) format. The maintenance fault tree includes fault classification, fault type, corresponding maintenance measures and parts recommendations, thus constructing a RAG knowledge base for the air conditioning maintenance field.

[0184] The large language model generates fault detection results. Utilizing a pre-trained large language model, based on user profile data and fault types, fault levels, and repair suggestions output from the RAG knowledge base, fault detection results are generated. The fault detection results include at least one repair plan, which may include detailed information such as required tools, replacement parts, repair steps, and estimated time, to suit at least one of the following: different user groups, different fault types, and different severity levels.

[0185] By utilizing recommendation system technology, fault detection results can be displayed to users and repairmen through mobile applications or smart home devices.

[0186] For faults that can be easily fixed with simple operations, the RAG knowledge base can output fault classifications that include both L1 and L2 levels. Then, using a large language model combined with user profile data for comprehensive analysis, the final fault classification is determined. This allows for dynamic adjustments to the L1 and L2 fault classifications by better incorporating user characteristics. For example, for a clogged filter, the knowledge base can output both L1 and L2 fault classifications. Based on user profile data, if the user profile indicates the user is a young or middle-aged male, the fault can be classified as L1, and self-repair instructions can be provided. If the user profile indicates the user is a middle-aged female or elderly person, the fault can be classified as L2, and a repair technician can be arranged for on-site repair.

[0187] Among them, the large language model can be pre-trained and supervised fine-tuned using a large Transformer-based model, and combined with multimodal data from millions of normal devices and multimodal data from 10,000 reported devices, as well as corresponding work order data, to improve the accuracy of fault prediction.

[0188] Continuous optimization of the large model. Online learning and incremental learning techniques can be used to achieve dynamic updates and adaptive learning of the model. By collecting maintenance feedback data, the expert mini-model and RAG knowledge base are continuously updated, as well as the large language model, enabling it to adapt to new environments and equipment states, thus improving the accuracy of fault prediction and the practicality of maintenance solutions. Compared to the static models of traditional methods, the continuous optimization and adaptive capabilities of this invention are more advantageous.

[0189] In some embodiments, a fault detection server is provided, such as Figure 10 As shown, the server includes:

[0190] Communication module 1002 is configured to establish a communication connection with at least one of an air conditioner and a user terminal;

[0191] Processor 1004 is configured as follows:

[0192] Upon receiving a fault report for an air conditioner, the system acquires the air conditioner's equipment operation data, including fault report data.

[0193] Based on fault description data, equipment operation data, and a pre-built knowledge base, detect the fault-related data and maintenance suggestion data corresponding to the air conditioner;

[0194] Based on fault correlation data, maintenance suggestion data, and user profile data corresponding to the air conditioner, fault detection results for the air conditioner are generated.

[0195] In some embodiments, the fault association data includes fault classification results; during the process of detecting fault association data and maintenance suggestion data corresponding to the air conditioner based on fault description data, equipment operation data, and a pre-built knowledge base, the processor is further configured to:

[0196] Based on the fault description data, the first initial classification result corresponding to the air conditioner is matched from the pre-set knowledge base, and the fault detection of the air conditioner is performed based on the equipment operation data to obtain the second initial classification result.

[0197] By combining the first initial classification result and the second initial classification result, the fault classification result is obtained;

[0198] Retrieve maintenance suggestion data corresponding to the fault classification results from the pre-built knowledge base.

[0199] In some embodiments, during the process of performing fault detection on the air conditioner based on device operating data to obtain a second initial classification result, the processor is further configured to:

[0200] Based on the pre-defined fault classification logic rules corresponding to various fault types, the fault type that matches the equipment operation data is detected, and a second initial classification result is obtained.

[0201] In some embodiments, during the process of fusing the first initial classification result and the second initial classification result to obtain the fault classification result, the processor is further configured to:

[0202] If the first initial classification result is inconsistent with the second initial classification result, and there is no confidence information in either the first or second initial classification result, the second initial classification result shall be determined as the fault classification result.

[0203] In some embodiments, the first initial classification result includes first confidence information corresponding to each of a plurality of first initial types; the second initial classification result includes second confidence information corresponding to each of a plurality of second initial types; each first initial type and each second initial type are at least partially consistent; in the process of fusing the first initial classification result and the second initial classification result to obtain the fault classification result, the processor is further configured to:

[0204] Based on the first confidence information and the second confidence information, the target confidence corresponding to each first initial type and each second initial type is detected;

[0205] The initial type corresponding to the highest target confidence level is determined as the fault classification result.

[0206] In some embodiments, during the process of detecting the target confidence corresponding to each first initial type and each second initial type based on each first confidence information and each second confidence information, the processor is further configured to:

[0207] Determine the target initial type from each of the first initial types and each of the second initial types;

[0208] Query the first confidence level corresponding to the initial type of the target from each first confidence level information, and query the second confidence level corresponding to the initial type of the target from each second confidence level information;

[0209] Based on the first weight corresponding to the first initial classification result and the second weight corresponding to the second initial classification result, the first confidence and the second confidence are weighted and fused to obtain the target confidence corresponding to the target initial type.

[0210] In some embodiments, the fault association data includes fault classification results and initial fault grading results; in the process of generating fault detection results for the air conditioner based on the fault association data, maintenance suggestion data, and user profile data corresponding to the air conditioner, the processor is further configured to:

[0211] When the initial fault classification result contains multiple alternative fault classification results, the target fault classification result is determined from each alternative fault classification result based on the user profile data corresponding to the air conditioner.

[0212] Based on the fault classification results, target fault grading results, and maintenance suggestion data, the fault detection results of the air conditioner are generated.

[0213] In some embodiments, before detecting fault-related data and maintenance suggestion data corresponding to the air conditioner based on fault description data, equipment operation data, and a pre-built knowledge base, the processor is further configured to:

[0214] Obtain at least one of the following in the field of air conditioners: fault code corresponding information, fault judgment mode data, historical repair work order data, fault repair measures information, and faulty component replacement information;

[0215] The fault code information, fault diagnosis mode data, historical repair work order data, fault repair measures information, and fault component replacement information are converted into tagged text data.

[0216] A knowledge base for the air conditioner domain is built based on tagged text data.

[0217] In some embodiments, the fault detection results are generated by a large language model; after generating the fault detection results for the air conditioner based on fault association data, maintenance suggestion data, and user profile data corresponding to the air conditioner, the processor is further configured to:

[0218] Upon receiving maintenance feedback data for air conditioners, the large language model is updated based on the maintenance feedback data.

[0219] In this embodiment, firstly, considering that fault description data has a certain degree of subjectivity and its accuracy depends on the user's observation and expression abilities, although it can intuitively and comprehensively reflect the abnormal performance of the air conditioner, and equipment operation data itself is more objective and accurate, it is also ambiguous, as different faults may produce the same or similar equipment operation data. Therefore, by combining subjective fault description data with objective equipment operation data, key information can be cross-validated, overcoming the limitations of a single data source, thereby significantly improving the accuracy and reliability of fault diagnosis. Secondly, by matching fault-related data and maintenance suggestion data from a pre-built knowledge base, a small amount of more relevant knowledge information can be filtered out and applied to fault detection. This not only effectively reduces the reasoning complexity of fault detection and improves fault detection efficiency, but also enhances the accuracy of fault detection. Thirdly, by combining user profiles for fault detection, more accurate fault detection results can be further inferred from the possible fault-related data and feasible maintenance suggestion data filtered from the knowledge base, making fault judgments that better match the user profile data and providing more suitable maintenance suggestions for the user. In this way, on the one hand, automated fault detection can reduce the time that maintenance personnel spend on fault detection and improve the efficiency of fault resolution; on the other hand, by improving the accuracy of automated fault detection, misjudgments can be reduced, thereby reducing the number of times maintenance personnel need to travel to the site and effectively improving the efficiency of fault resolution.

[0220] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the methods of the above embodiments.

[0221] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the methods of the above embodiments.

[0222] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the methods of the above embodiments.

[0223] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0224] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0225] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0226] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A fault detection method, characterized in that, Applied to a server, the method includes: Upon receiving a fault report for an air conditioner, the system acquires the equipment operation data of the air conditioner, wherein the fault report includes fault description data. Based on the fault description data, the equipment operation data, and the pre-set knowledge base, detect the fault association data and maintenance suggestion data corresponding to the air conditioner; Based on the fault association data, the maintenance suggestion data, and the user profile data corresponding to the air conditioner, the fault detection result of the air conditioner is generated.

2. The method according to claim 1, characterized in that, The fault association data includes fault classification results; the detection of fault association data and maintenance suggestion data corresponding to the air conditioner based on the fault description data, the equipment operation data, and the pre-set knowledge base includes: Based on the fault description data, the first initial classification result corresponding to the air conditioner is matched from the preset knowledge base, and the fault detection of the air conditioner is performed based on the equipment operation data to obtain the second initial classification result; By combining the first initial classification result and the second initial classification result, the fault classification result is obtained; Retrieve maintenance suggestion data corresponding to the fault classification results from the pre-built knowledge base.

3. The method according to claim 2, characterized in that, The process of fault detection of the air conditioner based on the equipment's operating data to obtain a second initial classification result includes: Based on the preset fault classification logic rules corresponding to various fault types, the fault type matching the equipment operation data is detected to obtain a second initial classification result.

4. The method according to claim 2, characterized in that, The fusion of the first initial classification result and the second initial classification result yields the fault classification result, including: If the first initial classification result is inconsistent with the second initial classification result, and there is no confidence information in either the first initial classification result or the second initial classification result, the second initial classification result shall be determined as the fault classification result.

5. The method according to claim 2, characterized in that, The first initial classification result includes first confidence information corresponding to each of the multiple first initial types; the second initial classification result includes second confidence information corresponding to each of the multiple second initial types; Each of the first initial types is at least partially identical to each of the second initial types; The fusion of the first initial classification result and the second initial classification result yields the fault classification result, including: Based on the first confidence information and the second confidence information, the target confidence corresponding to each of the first initial type and each of the second initial type is detected; The initial type corresponding to the highest target confidence level is determined as the fault classification result.

6. The method according to claim 5, characterized in that, The step of detecting the target confidence level corresponding to each of the first initial type and each of the second initial types based on the first confidence level information and the second confidence level information includes: Determine the target initial type from each of the first initial type and each of the second initial types; Query the first confidence level corresponding to the target initial type from each of the first confidence level information, and query the second confidence level corresponding to the target initial type from each of the second confidence level information; Based on the first weight corresponding to the first initial classification result and the second weight corresponding to the second initial classification result, the first confidence score and the second confidence score are weighted and fused to obtain the target confidence score corresponding to the target initial type.

7. The method according to claim 1, characterized in that, The fault association data includes fault classification results and initial fault grading results; the generation of fault detection results for the air conditioner based on the fault association data, the maintenance suggestion data, and the user profile data corresponding to the air conditioner includes: If the initial fault classification result includes multiple alternative fault classification results, the target fault classification result is determined from each of the alternative fault classification results based on the user profile data corresponding to the air conditioner. Based on the fault classification results, the target fault grading results, and the maintenance suggestion data, the fault detection results of the air conditioner are generated.

8. The method according to any one of claims 1 to 7, characterized in that, Before detecting the fault-related data and maintenance suggestion data corresponding to the air conditioner based on the fault description data, the equipment operation data, and the preset knowledge base, the method further includes: Obtain at least one of the following in the field of air conditioners: fault code corresponding information, fault judgment mode data, historical repair work order data, fault repair measures information, and faulty component replacement information; The fault code information, the fault judgment mode data, the historical maintenance work order data, the fault repair measures information, and the fault component replacement information are converted into tagged text data. Based on the tagged text data, a knowledge base for the air conditioner field is constructed.

9. The method according to any one of claims 1 to 7, characterized in that, The fault detection result is generated by a large language model; after generating the fault detection result of the air conditioner based on the fault association data, the maintenance suggestion data, and the user profile data corresponding to the air conditioner, the method further includes: Upon receiving maintenance feedback data for the air conditioner, the large language model is updated based on the maintenance feedback data.

10. A server, characterized in that, The server includes: The communication module is configured to establish a communication connection with at least one of the air conditioner and the user terminal; The processor is configured as follows: Upon receiving a fault report for an air conditioner, the system acquires the equipment operation data of the air conditioner, wherein the fault report includes fault description data. Based on the fault description data, the equipment operation data, and the pre-set knowledge base, detect the fault association data and maintenance suggestion data corresponding to the air conditioner; Based on the fault association data, the maintenance suggestion data, and the user profile data corresponding to the air conditioner, the fault detection result of the air conditioner is generated.

Images