Fault diagnosis method, diagnosis device and equipment for range hood and medium

Abstract

The embodiment of the invention discloses a fault diagnosis method, diagnosis device, equipment and medium for a range hood, and the method comprises the steps: firstly, synchronously collecting an acoustic signal, a vibration signal and a working condition signal during the operation of the range hood, and carrying out the data processing; respectively obtaining an acoustic feature time sequence corresponding to the acoustic signal, a vibration feature time sequence corresponding to the vibration signal and a working condition feature time sequence corresponding to the working condition signal; performing data fusion processing on the acoustic feature sequential sequence, the vibration feature sequential sequence and the working condition feature sequential sequence to obtain a fused feature vector, then performing similarity matching analysis on the fused feature vector and a plurality of standard feature vectors in a preset knowledge graph to correspondingly obtain a plurality of similarity calculated values, and calculating the similarity calculated values according to the similarity calculated values. And finally, according to the plurality of similarity calculation values, determining a fault diagnosis result of the extractor hood. According to the method, the monitoring precision of the operation condition of the range hood and the judgment of system abnormity can be improved.

Description

Technical Field

[0001] This invention relates to the field of household appliance technology, and in particular to a fault diagnosis method, diagnostic device, equipment and medium for a range hood. Background Technology

[0002] Range hoods have become an indispensable kitchen appliance in modern homes. They operate on the principles of fluid dynamics, using a centrifugal fan inside to draw in cooking fumes and a filter to remove some grease particles. The centrifugal fan consists of a casing, an impeller housed within the casing, and a motor that drives the impeller. As the impeller rotates, a negative pressure is generated at the center of the fan, drawing in the cooking fumes from below. After being accelerated by the fan, the fumes are collected by the casing and guided outdoors. The centrifugal fan is also the power source of the range hood.

[0003] Traditional methods for detecting malfunctions in range hoods typically involve collecting and analyzing vibration signals from the motor during operation. This allows for real-time monitoring of the range hood's operating status, providing timely reminders to users to clean or replace the impeller to prevent wear on the internal structure of the volute. However, existing methods for detecting malfunctions in range hoods have the following problems:

[0004] (1) Single perception: relying solely on vibration signals, the information dimensions are insufficient, making it difficult to distinguish fault types (such as being unable to distinguish between impeller imbalance and bearing damage), resulting in low anomaly detection accuracy;

[0005] (2) Isolated diagnosis: Single-machine operation, unable to utilize historical big data and collaborative information for cross-validation, resulting in a low upper limit of diagnostic accuracy;

[0006] (3) Lack of knowledge accumulation: Diagnostic rules are fixed and cannot evolve with the accumulation of data, lacking the ability to build and reuse fault knowledge base;

[0007] (4) Inflexible response to false alarms: It can only filter occasional interference through simple delay or threshold, and has poor anti-interference ability against unknown and complex composite interference sources (such as pots falling or knocking on the door). Summary of the Invention

[0008] This invention provides a fault diagnosis method, diagnostic device, equipment, and medium for range hoods. It uses an acoustic-vibration fusion method based on multimodal fusion analysis of acoustic and vibration signals to detect anomalies in range hoods, and utilizes cloud-based knowledge graphs for remote fault diagnosis of range hoods. This improves the monitoring accuracy of range hood operating conditions and enhances the judgment and diagnostic accuracy of range hood system anomalies.

[0009] In a first aspect, embodiments of the present invention provide a method for diagnosing faults in a range hood, comprising:

[0010] The acoustic signals, vibration signals, and operating condition signals of the range hood are collected synchronously during operation and processed to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the operating condition feature time sequence corresponding to the operating condition signal, respectively.

[0011] A multimodal fusion deep learning model based on an attention mechanism is used to perform data fusion processing on the acoustic feature time sequence, the vibration feature time sequence, and the working condition feature time sequence to obtain a fused feature vector;

[0012] The fused feature vector is compared with multiple standard feature vectors in a preset knowledge graph for similarity matching analysis, resulting in multiple similarity calculation values.

[0013] The fault diagnosis result of the range hood is determined based on multiple similarity calculation values.

[0014] Secondly, embodiments of the present invention also provide a fault diagnosis device for a range hood, comprising:

[0015] The data acquisition and processing module is used to synchronously acquire acoustic signals, vibration signals and operating condition signals during the operation of the range hood, and perform data processing to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal and the operating condition feature time sequence corresponding to the operating condition signal, respectively.

[0016] The data fusion module is used to perform data fusion processing on the acoustic feature time series, the vibration feature time series and the working condition feature time series using a multimodal fusion deep learning model based on an attention mechanism to obtain a fused feature vector;

[0017] The similarity calculation module is used to perform similarity matching analysis between the fused feature vector and multiple standard feature vectors in the preset knowledge graph, and obtain multiple similarity calculation values ​​accordingly.

[0018] The fault diagnosis module is used to determine the fault diagnosis result of the range hood based on multiple similarity calculation values.

[0019] Thirdly, embodiments of the present invention also provide a terminal device, including:

[0020] One or more processors;

[0021] Storage device for storing one or more programs;

[0022] When the one or more programs are executed by the one or more processors, the one or more processors implement the fault diagnosis method for the range hood as described in any of the first aspects.

[0023] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the fault diagnosis method for a range hood as described in any of the first aspects.

[0024] This invention provides a fault diagnosis method, diagnostic device, equipment, and medium for a range hood. The fault diagnosis method first simultaneously collects acoustic signals, vibration signals, and operating condition signals during the operation of the range hood and processes the data to obtain acoustic feature time-series sequences corresponding to the acoustic signals, vibration feature time-series sequences corresponding to the vibration signals, and operating condition feature time-series sequences corresponding to the operating condition signals. Then, a multimodal fusion deep learning model based on an attention mechanism is used to fuse the acoustic feature time-series sequences, vibration feature time-series sequences, and operating condition feature time-series sequences to obtain a fused feature vector. Next, the fused feature vector is compared with multiple standard feature vectors in a preset knowledge graph for similarity matching analysis, resulting in multiple similarity calculation values. Finally, based on these multiple similarity calculation values, the fault diagnosis result of the range hood is determined. Using the above methods, acoustic signals, vibration signals, and operating condition signals of the range hood are collected during operation. Acquiring data from multiple information dimensions ensures further detailed analysis for fault diagnosis. A multimodal fusion deep learning model based on an attention mechanism is employed to fuse the acoustic feature time-series sequences corresponding to the acoustic signals, the vibration feature time-series sequences corresponding to the vibration signals, and the operating condition feature time-series sequences corresponding to the operating condition signals. This allows for subsequent fault diagnosis of the range hood based on similarity matching analysis of the fused feature vectors. In other words, anomaly detection of the range hood can be performed using an acoustic-vibration fusion method based on multimodal fusion analysis of acoustic and vibration signals, and the range hood can be further analyzed using a pre-set knowledge graph in the cloud. Remote diagnosis of range hood malfunctions avoids the problems of difficulty in distinguishing fault types and low accuracy of fault diagnosis results caused by single or insufficient data collection dimensions. It also avoids the problem of the inability to cross-verify historical big data and collaborative information due to single-machine computing. It effectively improves the monitoring accuracy of range hood operating conditions and enhances the judgment and diagnosis accuracy of range hood system anomalies. The system can evolve with the accumulation of data from multiple fault diagnosis results, generating knowledge accumulation. This makes the content of the cloud-preset knowledge graph more consistent with the anomaly detection process of the range hood. Multiple information dimensions and threshold analysis and comparison are used to filter out occasional interference, and the anti-interference ability against unknown and complex compound interference sources is also enhanced.

[0025] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating a fault diagnosis method for a range hood provided in an embodiment of the present invention;

[0028] Figure 2 This is a flowchart illustrating another method for diagnosing a range hood fault provided in an embodiment of the present invention;

[0029] Figure 3 This is a flowchart illustrating another method for diagnosing a range hood fault provided in an embodiment of the present invention;

[0030] Figure 4 This is a flowchart illustrating another method for diagnosing a range hood fault provided in an embodiment of the present invention;

[0031] Figure 5 This is a flowchart illustrating another method for diagnosing a range hood fault provided in an embodiment of the present invention;

[0032] Figure 6 This is a flowchart illustrating another method for diagnosing a range hood fault provided in an embodiment of the present invention;

[0033] Figure 7 This is a schematic diagram of the structure of a fault diagnosis device for a range hood provided in an embodiment of the present invention;

[0034] Figure 8 This is a schematic diagram of the structure of a terminal device provided in an embodiment of the present invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0036] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0037] Figure 1 This is a flowchart illustrating a fault diagnosis method for a range hood provided in an embodiment of the present invention. This fault diagnosis method is applicable to detecting abnormal states during the operation of a range hood. The method can be executed by a fault diagnosis device for the range hood, which can be implemented in hardware and / or software and can be configured in a control board. Figure 1 As shown, the fault diagnosis method includes:

[0038] S110: Synchronously acquire acoustic signals, vibration signals, and operating condition signals during the operation of the range hood, and perform data processing to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the operating condition feature time sequence corresponding to the operating condition signal, respectively.

[0039] Specifically, this step essentially involves collecting signals from multiple dimensions during the operation of the range hood and processing these signals to meet the input requirements of a multimodal fusion deep learning model based on an attention mechanism. This allows for subsequent comprehensive analysis of the multiple signals during the range hood's operation to obtain fault diagnosis results. For example, a built-in microphone array can be used to capture wide-band operating noise and collect acoustic signals during range hood operation; these acoustic signals can also be understood as audio signals. For example, a vibration sensor can be used to collect vibration signals during range hood operation; this vibration sensor may include a triaxial accelerometer. For example, a speed sensor can be used to collect the motor speed signal during range hood operation and use this motor speed signal as the operating condition signal of the range hood. Furthermore, the operating condition signal of the range hood may also include, but is not limited to, motor current, motor power, and the opening degree of the smoke baffle; appropriate signal acquisition equipment can be selected according to actual needs. By using multimodal signal synchronous acquisition and a hardware synchronization mechanism, it can be ensured that the acoustic signals, vibration signals and operating condition signals of the range hood are acquired at the same time interval. In other words, it ensures that the data acquisition of the acoustic signals, vibration signals and operating condition signals of the range hood has a unified time, thus achieving precise time-frequency alignment.

[0040] Furthermore, the acoustic feature time series corresponding to the acoustic signal may include, but is not limited to, sound intensity features, which can map the feature information of the acoustic signal to the corresponding acquisition time. The vibration feature time series corresponding to the vibration signal may include, but is not limited to, time-domain features and frequency-domain features, which can map the feature information of the vibration signal to the corresponding acquisition time. The operating condition feature time series corresponding to the operating condition signal may include, but is not limited to, rotational speed features, which can map the feature information of the operating condition signal to the corresponding acquisition time.

[0041] S120. A multimodal fusion deep learning model based on attention mechanism is used to perform data fusion processing on acoustic feature time series, vibration feature time series and working condition feature time series to obtain fused feature vector.

[0042] Specifically, this step involves inputting the acoustic feature time series, vibration feature time series, and operating condition feature time series into a multimodal fusion deep learning model based on an attention mechanism. This model performs data fusion processing on these feature time series and outputs a fused feature vector, which can then be used for subsequent fault diagnosis. The attention-based multimodal fusion deep learning model can perform fusion analysis and deep learning, aggregating massive amounts of equipment data and utilizing powerful computing power and algorithm models to perform complex correlation analysis, pattern recognition, and intelligent decision-making, forming "global intelligence."

[0043] For example, a multimodal fusion deep learning model based on an attention mechanism may include a cloud access layer and a signal reconstruction and alignment layer. The cloud access layer can receive concurrent data connections from millions of devices and perform efficient scheduling and preprocessing. The signal reconstruction and alignment layer can reassemble the received acoustic feature time series, vibration feature time series, and operating condition feature time series to reconstruct the specific scenario in which the data was generated, providing a complete data package for subsequent analysis and ensuring that the acoustic feature time series and vibration feature time series each carry contextual information such as their corresponding operating condition feature time series and acquisition time. The multimodal fusion diagnostic engine can then comprehensively summarize this information. Alternatively, for example, a multimodal fusion deep learning model based on an attention mechanism is a complete, end-to-end system architecture that may include an acoustic analysis subnetwork, a vibration analysis subnetwork, and an attention fusion module. The acoustic analysis subnetwork is a deep neural network specifically responsible for learning patterns from the temporal sequence of acoustic features and identifying abnormal sound patterns such as howling, friction, and imbalance. The vibration analysis subnetwork is another deep neural network specifically responsible for learning patterns from the temporal sequence of vibration features and identifying abnormal vibration patterns such as impact, resonance, and eccentricity. The attention fusion module is the "brain" of the entire system, which integrates and summarizes information to dynamically and intelligently determine which sensor and feature are most useful for diagnosing the current fault at a given moment. It should also be noted that the temporal sequences of acoustic features, vibration features, and operating conditions are input into the corresponding deep neural networks. These networks first learn the temporal patterns within each mode and output a generalized feature vector. Then, the attention mechanism acts like a "smart switch," dynamically determining how to combine the feature vectors of these different modes based on the current operating conditions (such as rotational speed) and the importance of the features themselves. In a more easily understood way, the acoustic analysis subnetwork and vibration analysis subnetwork can be responsible for stating, "Signs of bearing failure have been found in the vibration signal," while the attention fusion module can be responsible for stating, "The acoustic evidence is more reliable than the vibration evidence, so I should trust the acoustic analysis results more." The acoustic feature time series, vibration feature time series, and operating condition feature time series can be input into the corresponding attention-based multimodal fusion deep learning model, instead of directly inputting the collected acoustic signals, vibration signals, and operating condition signals into the corresponding attention-based multimodal fusion deep learning model. The resulting fusion feature vector is a highly abstract representation; its dimensions no longer directly correspond to specific vibration signals, acoustic signals, etc., but represent a comprehensive fault probability distribution, which can be understood as a "weighted comprehensive diagnostic report."

[0044] S130. Perform similarity matching analysis between the fused feature vector and multiple standard feature vectors in the preset knowledge graph to obtain multiple similarity calculation values.

[0045] Specifically, this step is essentially a process of querying and comparing a pre-defined knowledge graph. This pre-defined knowledge graph structures expert knowledge, historical repair records, and component relationships, making the final fault diagnosis result not just a label, but an interpretable and reasonable solution. Each symptom feature node (or fault mode) in the pre-defined knowledge graph has a corresponding standard feature vector. The fused feature vector is then matched with these standard feature vectors to perform similarity analysis, thereby accurately determining the fault diagnosis result of the range hood. For example, the symptom feature nodes (or fault modes) in the pre-defined knowledge graph may include, but are not limited to, impeller imbalance, bearing wear, and motor eccentricity.

[0046] S140. Determine the fault diagnosis result of the range hood based on multiple similarity calculation values.

[0047] Specifically, this step is essentially a decision-making process based on a pre-defined knowledge graph. Each symptom feature node (or fault mode) in the pre-defined knowledge graph has a corresponding standard feature vector. After performing similarity matching analysis between the fused feature vector and multiple standard feature vectors in the pre-defined knowledge graph, the standard feature vector that best matches the fused feature vector can be found in the pre-defined knowledge graph. Therefore, the symptom feature node (or fault mode) corresponding to this standard feature vector is likely the cause of the range hood's malfunction, and this symptom feature node (or fault mode) can be displayed in the range hood's fault diagnosis results. For example, based on the obtained multiple similarity calculation values, the pre-defined knowledge graph can be traversed to finally lock in the most likely symptom feature node (or fault mode) and return its confidence score (or matching value, similarity calculation value). The confidence score (or matching value, similarity calculation value) can be understood as the probability that the symptom feature node (or fault mode) is the actual cause of the fault. For example, the fault diagnosis results of a range hood can ultimately be used to generate a fault diagnosis report that users can understand and that after-sales service can execute.

[0048] Furthermore, the decision-making process based on the pre-defined knowledge graph not only provides fault diagnosis results for the range hood's anomalies but also accurately identifies the specific location of the anomaly and recommends necessary replacement parts, facilitating model evolution. For example, the user reminder and after-sales system translates the intelligent fault diagnosis results into actionable instructions, alerting users, such as sending a friendly notification via the app saying, "Your range hood may need maintenance." For example, the fault diagnosis results can also be linked to the after-sales system to automatically generate repair work orders and even pre-dispatch expected parts to nearby engineers, greatly improving service efficiency and quality. For example, after on-site repairs, after-sales personnel can confirm the cause of the fault via the app; this fault diagnosis result serves as a labeled sample fed back to the cloud system for model evolution. For example, the system uses new samples for incremental learning of the model, continuously optimizing model accuracy and the completeness of the pre-defined knowledge graph. For example, regarding the handling of "false alarms" and "instability", such as the unstable speed stage, by analyzing the transient response characteristics of sound and vibration at the moment of gear shift, problems such as inertial imbalance and motor starting failure can be detected early; for example, regarding occasional interference, the model has a very strong anti-interference capability. The system will judge the duration and correlation of abnormal modes. Occasional interference is usually not synchronized with the rotation frequency and has a short duration, so it will be filtered out by the model.

[0049] In one specific implementation, the process of determining the final fault diagnosis result using a multimodal fusion deep learning model based on an attention mechanism and a preset knowledge graph is demonstrated for a specific fault scenario. For example, the fault scenario could be a user reporting excessive vibration and abnormal noise from a range hood. First, the cloud-based multimodal fusion deep learning model based on an attention mechanism performs preliminary detection and analysis, outputting a fusion feature vector. This fusion feature vector reflects characteristics such as high-frequency vibration energy and prominent acoustic harmonics at specific frequencies. At this point, the initial judgment is that it is a bearing-related problem, with a confidence level (or matching value, similarity calculation value) of 85%. Then, the preset knowledge graph intervenes, moving from "what" to "why" and "how." After obtaining the preliminary judgment of the bearing-related problem, the following reasoning process can be automatically executed in the preset knowledge graph.

[0050] Question 1: What specific symptoms are usually associated with "bearing wear"?

[0051] The preset knowledge graph query found: high-frequency vibration (kurtosis > 4), sound spectrum has harmonics at specific frequencies (such as 3.5kHz).

[0052] Interpretable output (which can be displayed in the fault diagnosis results): Comparing the current data "Vibration kurtosis of 5.2 was detected, and there are significant harmonics at 3.6kHz", this is highly consistent with the typical symptoms of bearing wear.

[0053] Question 2: What is the root cause of "bearing wear"?

[0054] The preset knowledge graph query found the following: lubrication failure (70%), foreign object ingress (20%), and normal aging (10%).

[0055] Interpretable output (which can be displayed in the fault diagnosis results): Combined with the equipment information "Your equipment has been running for 2 years, exceeding the normal lubrication cycle", the preliminary judgment is that lubrication failure caused bearing wear.

[0056] Query 3: What specific operations and parts are needed to solve this problem?

[0057] The preset knowledge graph query found that the operation is to replace the bearing, the required part is the bearing, and the related component is the oil seal.

[0058] Interpretable output (which can be displayed in the fault diagnosis results): The solution is to replace the bearing, the required bearing part number is XX, and it is recommended to check and replace the adjacent oil seal at the same time, the required oil seal part number is XX.

[0059] Question 4: What are the consequences if this question is ignored?

[0060] The preset knowledge graph query found that: increased vibration will further lead to wear on the motor shaft, and further lead to motor burnout.

[0061] Interpretable output (which can be displayed in the fault diagnosis results): Risk warning: If not handled in time, it may cause motor shaft wear within one month, increasing maintenance costs by 300%.

[0062] Finally, a definitive, interpretable, and reasonable fault diagnosis report is generated. Based on the query and reasoning of the above-preset knowledge graph, the final fault diagnosis result will not simply provide a "bearing failure" label, but will generate a comprehensive diagnostic report. Table 1 is a schematic table of a fault diagnosis report provided by an embodiment of the present invention, which can be referred to as Table 1 below.

[0063] Table 1

[0064]

[0065] The technical solution in this invention collects acoustic signals, vibration signals, and operating condition signals during the operation of the range hood. Acquiring data from multiple information dimensions ensures further detailed analysis for fault diagnosis. A multimodal fusion deep learning model based on an attention mechanism is employed to fuse the acoustic feature time-series sequences corresponding to the acoustic signals, the vibration feature time-series sequences corresponding to the vibration signals, and the operating condition feature time-series sequences corresponding to the operating condition signals. This facilitates subsequent fault diagnosis of the range hood based on similarity matching analysis of the fused feature vectors. In other words, anomaly detection of the range hood can be performed using an acoustic-vibration fusion method based on multimodal fusion analysis of acoustic and vibration signals. Furthermore, remote fault diagnosis of the range hood can be performed using a pre-set knowledge graph in the cloud. This avoids the problem of cross-validation of historical big data and collaborative information caused by single-machine computation, effectively improving the monitoring accuracy of the range hood's operating condition and enhancing the judgment and diagnostic accuracy of the range hood system anomalies. The system can evolve with the accumulation of data from multiple fault diagnosis results, making the content of the pre-set knowledge graph in the cloud more consistent with the anomaly detection process of the range hood, thus enhancing its anti-interference capability.

[0066] Figure 2 This is a flowchart illustrating another fault diagnosis method for a range hood provided in this embodiment of the invention. This embodiment is an optimization based on the above embodiment. Optionally, acoustic signals, vibration signals, and operating condition signals during the operation of the range hood are simultaneously collected and processed to obtain the acoustic feature time sequence corresponding to the acoustic signals, the vibration feature time sequence corresponding to the vibration signals, and the operating condition feature time sequence corresponding to the operating condition signals, including:

[0067] Simultaneously collect acoustic signals, vibration signals, and operating condition signals during the operation of the range hood;

[0068] Feature extraction and data compression are performed on acoustic signals, vibration signals, operating condition signals, and their corresponding acquisition times to obtain feature vector data packets.

[0069] In the time dimension, the feature vector data packets are reconstructed to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the operating condition feature time sequence corresponding to the operating condition signal.

[0070] For details not covered in this embodiment, please refer to the above embodiments. Figure 2 As shown, the fault diagnosis method includes:

[0071] S210: Synchronously collect acoustic signals, vibration signals, and operating condition signals during the operation of the range hood.

[0072] Optionally, acoustic signals, vibration signals, and operating condition signals of the range hood are collected simultaneously, including: at the same time, using a microphone array to collect acoustic signals, using a vibration sensor to collect vibration signals, and using a speed sensor to collect operating condition signals.

[0073] For example, at the same acquisition time, a built-in microphone array can be used to capture wide-band operating noise and acquire the acoustic signal of the range hood during operation; a vibration sensor can be used to acquire the vibration signal of the range hood during operation; and a speed sensor can be used to acquire the motor speed signal of the range hood during operation. This motor speed signal is then used as the operating condition signal of the range hood. In this way, the acquisition of the acoustic signal, vibration signal, and operating condition signal of the range hood during operation is ensured to be at a unified time, achieving precise time-frequency alignment.

[0074] S220. Perform feature extraction and data compression processing on acoustic signals, vibration signals, operating condition signals and corresponding acquisition times to obtain feature vector data packets.

[0075] Specifically, preliminary processing of acoustic signals, vibration signals, and operating condition signals significantly reduces data transmission volume. Feature vectors obtained after feature extraction from acoustic and vibration signals are packaged with synchronized operating condition signals into a lightweight data packet. The feature vector obtained after feature extraction can be understood as a set of multiple features. For example, three features can be extracted from a vibration signal: root mean square (RMS, representing energy), kurtosis (representing impact), and center-of-gravity frequency (representing frequency distribution). These three feature values ​​can then form a three-dimensional vector, which is the feature vector corresponding to the vibration signal feature extraction. Feature extraction and data compression are used to achieve high-quality signal acquisition and preliminary information condensation at the data generation source, preparing high-quality, low-redundancy information for cloud-based decision-making. Massive amounts of raw data are "refined" into high-value, smaller "feature vectors," compressing the original several MB of raw data per second into "feature vectors" requiring only a few KB per second, achieving data dimensionality reduction.

[0076] S230. In the time dimension, the feature vector data packet is reconstructed to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the working condition feature time sequence corresponding to the working condition signal.

[0077] For example, the processed feature vector data packets can be securely, reliably, and efficiently uploaded to the cloud via a wireless communication module. Specifically, after receiving the feature vector data packets, the cloud can reconstruct the feature time-series sequences of each modality, such as the acoustic feature time-series sequence corresponding to acoustic signals, the vibration feature time-series sequence corresponding to vibration signals, and the operating condition feature time-series sequence corresponding to operating condition signals. Data reconstruction processing aims to restore the temporal structure of the data so that the cloud model can use information in the time dimension for fault diagnosis. This can also be understood as mapping the feature information of acoustic signals, vibration signals, and operating condition signals to their corresponding acquisition times. Furthermore, it should be noted that the feature vector data packets before data reconstruction processing are compressed and serialized to save bandwidth but cannot be directly used for analysis. The acoustic feature time-series, vibration feature time-series, and operating condition feature time-series sequences after data reconstruction processing are decompressed and arranged according to time steps, and can be directly input into a multimodal fusion deep learning model based on an attention mechanism for analysis.

[0078] S240. A multimodal fusion deep learning model based on attention mechanism is used to perform data fusion processing on acoustic feature time series, vibration feature time series and working condition feature time series to obtain fused feature vector.

[0079] S250. Perform similarity matching analysis between the fused feature vector and multiple standard feature vectors in the preset knowledge graph to obtain multiple similarity calculation values.

[0080] S260. Based on multiple similarity calculation values, determine the fault diagnosis result of the range hood.

[0081] Figure 3 This is a flowchart illustrating another fault diagnosis method for a range hood provided by an embodiment of the present invention. This embodiment is an optimization based on the above embodiment. Optionally, a multimodal fusion deep learning model based on an attention mechanism is used to perform data fusion processing on the acoustic feature time series sequence, vibration feature time series sequence, and operating condition feature time series sequence to obtain a fused feature vector, including:

[0082] Based on the attention-based multimodal fusion deep learning model, the weights corresponding to the acoustic feature time series, vibration feature time series, and working condition feature time series are determined respectively.

[0083] Based on the weights corresponding to the acoustic feature time series, vibration feature time series, and operating condition feature time series, a weighted fusion process is performed on the acoustic feature time series, vibration feature time series, and operating condition feature time series to obtain a fused feature vector.

[0084] For details not covered in this embodiment, please refer to the above embodiments. Figure 3As shown, the fault diagnosis method includes:

[0085] S310: Synchronously acquire acoustic signals, vibration signals, and operating condition signals during the operation of the range hood, and perform data processing to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the operating condition feature time sequence corresponding to the operating condition signal.

[0086] S320. Based on the multimodal fusion deep learning model based on the attention mechanism, determine the weights corresponding to the acoustic feature time series, vibration feature time series, and working condition feature time series, respectively.

[0087] Specifically, the attention mechanism can automatically calculate the importance weight of each modal feature (such as acoustic feature time series, vibration feature time series and operating condition feature time series) at each time step, which can also be understood as attention score.

[0088] For example, suppose a range hood is experiencing early bearing wear. This type of fault generates high-frequency vibration signals, which might manifest as a high-frequency hissing sound in the acoustic signal. However, due to the noisy kitchen environment, the acoustic signal may be interfered with, while the vibration signal is relatively reliable. In a multimodal fusion deep learning model based on an attention mechanism, the vibration feature time series corresponding to the vibration signal would be assigned a higher weight because the vibration signal is more sensitive and reliable for bearing faults. Simultaneously, the model will also appropriately consider the high-frequency components in the acoustic feature time series corresponding to the sound signal, but with a lower weight. Alternatively, if the fault is caused by foreign objects adhering to the impeller, leading to increased aerodynamic noise, the acoustic signal might more clearly reflect this change. In this case, the multimodal fusion deep learning model based on an attention mechanism would assign a higher weight to the acoustic feature time series corresponding to the acoustic signal. In this way, the multimodal fusion deep learning model based on an attention mechanism can adaptively select the most relevant information, improving the accuracy of diagnosis.

[0089] S330. Based on the weights corresponding to the acoustic feature time series, vibration feature time series, and operating condition feature time series, perform weighted fusion processing on the acoustic feature time series, vibration feature time series, and operating condition feature time series to obtain the fused feature vector.

[0090] Specifically, acoustic feature time series, vibration feature time series, and operating condition feature time series can be input into a multimodal fusion deep learning model based on an attention mechanism. These feature time series can be fused according to their respective weights, and a fused feature vector can be output so that subsequent fault diagnosis can be performed based on the fused feature vector.

[0091] S340. Perform similarity matching analysis between the fused feature vector and multiple standard feature vectors in the preset knowledge graph to obtain multiple similarity calculation values.

[0092] S350. Based on multiple similarity calculation values, determine the fault diagnosis result of the range hood.

[0093] Figure 4 This is a flowchart illustrating another fault diagnosis method for a range hood provided by an embodiment of the present invention. This embodiment is an optimization based on the above embodiment. Optionally, the fused feature vector is compared with multiple standard feature vectors in a preset knowledge graph for similarity matching analysis, resulting in multiple similarity calculation values, including:

[0094] Based on the principle of cosine similarity, the fused feature vector is compared with each standard feature vector to calculate the similarity, resulting in multiple similarity values.

[0095] For details not covered in this embodiment, please refer to the above embodiments. Figure 4 As shown, the fault diagnosis method includes:

[0096] S410 synchronously acquires acoustic signals, vibration signals, and operating condition signals during the operation of the range hood, and performs data processing to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the operating condition feature time sequence corresponding to the operating condition signal.

[0097] S420. A multimodal fusion deep learning model based on attention mechanism is used to perform data fusion processing on acoustic feature time series, vibration feature time series and working condition feature time series to obtain fused feature vector.

[0098] S430. Based on the cosine similarity principle, the fused feature vector is compared with each standard feature vector to calculate the similarity, resulting in multiple similarity values.

[0099] Specifically, each symptom feature node (or fault mode) in the preset knowledge graph has a corresponding standard feature vector. Table 2 is a schematic table illustrating the correspondence between symptom feature nodes (fault modes) and their corresponding standard feature vectors provided in this embodiment of the invention. This is merely an example and not intended to limit the scope. For instance, the fusion feature vector output by the multimodal fusion deep learning model based on the attention mechanism can be [0.92, 0.15, 0.85, 0.03]. Similarity can then be calculated based on the cosine similarity principle. The cosine similarity principle can also be understood as a mathematical matching process; it does not compare the absolute size of the vectors, but rather compares whether their directions are consistent. The closer the calculated value is to 1, the more similar the patterns of the two vectors are. The cosine similarity principle calculates similarity using the formula: Similarity value = (fused feature vector • standard feature vector) / (||fused feature vector|| × ||standard feature vector||), where (fused feature vector • standard feature vector) can be understood as the vector dot product, and (||fused feature vector|| × ||standard feature vector||) can be understood as the product of vector magnitudes. Furthermore, by way of example, other similarity calculation methods can also be used to process the similarity between the fused feature vector and each standard feature vector to find the standard feature vector that is most similar to the fused feature vector.

[0100] Table 2

[0101]

[0102] For example, after performing similarity calculations between the fused feature vector and the standard feature vector corresponding to "bearing inner ring wear", the vector dot product = 0.92 × 0.95 + 0.15 × 0.10 + 0.85 × 0.88 + 0.03 × 0.02 = 1.6376, and the vector magnitude product = × =1.664, then the similarity calculation value between the fused feature vector and the standard feature vector corresponding to "bearing inner ring wear" is 1.6376 / 1.664 ≈ 0.984. Similarly, when performing similarity calculation between the fused feature vector and the standard feature vector corresponding to "impeller imbalance", the vector dot product is 0.54 and the vector magnitude product is 1.222, so the similarity calculation value between the fused feature vector and the standard feature vector corresponding to "impeller imbalance" is 0.54 / 1.222 ≈ 0.442. Similarly, when performing similarity calculation between the fused feature vector and the standard feature vector corresponding to "bearing lubrication deficiency", the vector dot product is 1.1843 and the vector magnitude product is 1.207, so the similarity calculation value between the fused feature vector and the standard feature vector corresponding to "bearing lubrication deficiency" is 1.1843 / 1.207 ≈ 0.981. Similarly, the similarity calculation is performed between the fused feature vector and the standard feature vector corresponding to "motor loose". The vector dot product is 0.9445 and the vector magnitude product is 1.620. Therefore, the similarity calculation value obtained by the fused feature vector and the standard feature vector corresponding to "motor loose" is 0.9445 / 1.620≈0.583.

[0103] S440. Based on multiple similarity calculation values, determine the fault diagnosis result of the range hood.

[0104] Figure 5 This is a flowchart illustrating another method for diagnosing a range hood's faults according to an embodiment of the present invention. This embodiment is an optimization based on the above embodiments. Optionally, the fault diagnosis result of the range hood is determined based on multiple similarity calculation values, including:

[0105] Among multiple similarity calculation values, the similarity calculation value that satisfies the first condition is determined; wherein, the first condition is that it is greater than the similarity calculation threshold and is the maximum value among the similarity calculation values;

[0106] Based on the standard feature vectors corresponding to the similarity calculation values ​​that meet the first condition, and the preset mapping relationship between the standard feature vectors and the fault modes, the determined fault modes are used as the matching fault diagnosis results of the range hood.

[0107] For details not covered in this embodiment, please refer to the above embodiments. Figure 5 As shown, the fault diagnosis method includes:

[0108] S510 synchronously acquires acoustic signals, vibration signals, and operating condition signals during the operation of the range hood, and performs data processing to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the operating condition feature time sequence corresponding to the operating condition signal.

[0109] S520. A multimodal fusion deep learning model based on attention mechanism is used to perform data fusion processing on acoustic feature time series, vibration feature time series and working condition feature time series to obtain fused feature vector.

[0110] S530. Perform similarity matching analysis between the fused feature vector and multiple standard feature vectors in the preset knowledge graph to obtain multiple similarity calculation values.

[0111] S540. Among multiple similarity calculation values, determine the similarity calculation value that satisfies the first condition; wherein, the first condition is that it is greater than the similarity calculation threshold and is the maximum value among the similarity calculation values.

[0112] S550. Based on the similarity calculation value that meets the first condition, the corresponding standard feature vector to be calculated, and the preset mapping relationship between the standard feature vector and the fault mode, the determined fault mode is used as the matching fault diagnosis result of the range hood.

[0113] Specifically, this step essentially involves making a diagnostic decision based on the matching results of multiple similarity calculations. Among these similarity calculations, the maximum value is identified, and the standard feature vector corresponding to this maximum value is determined. Then, based on the relationships within a pre-defined knowledge graph, the fault mode (or symptom feature node) corresponding to this standard feature vector is determined. The fault mode corresponding to this standard feature vector can then be used as the matching fault diagnosis result for the range hood.

[0114] For example, continuing to refer to Table 2, the similarity calculated between the fused feature vector and the standard feature vector corresponding to "bearing inner ring wear" is 0.984, the similarity calculated between the fused feature vector and the standard feature vector corresponding to "bearing oil shortage" is 0.981, the similarity calculated between the fused feature vector and the standard feature vector corresponding to "motor looseness" is 0.583, and the similarity calculated between the fused feature vector and the standard feature vector corresponding to "impeller imbalance" is 0.442. The similarity calculation threshold can be 0.95. Since the similarity calculated between the fused feature vector and the standard feature vector corresponding to "bearing inner ring wear" is the maximum value and greater than the similarity calculation threshold, the matching fault diagnosis result for the range hood can be generated based on the bearing inner ring wear corresponding to 0.984.

[0115] In another specific embodiment, optionally, determining the fault diagnosis result of the range hood based on multiple similarity calculation values ​​further includes: determining the similarity calculation value that satisfies a second condition among the multiple similarity calculation values; wherein the second condition is greater than a similarity calculation threshold and less than the maximum value among the similarity calculation values; and using the determined fault mode as a candidate fault diagnosis result of the range hood based on the standard feature vector corresponding to the similarity calculation value that satisfies the second condition and the preset mapping relationship between the standard feature vector and the fault mode.

[0116] For example, continuing to refer to Table 2, the similarity calculation values ​​obtained by fusing the feature vector with the standard feature vector corresponding to "bearing inner ring wear" and the similarity calculation values ​​obtained by fusing the feature vector with the standard feature vector corresponding to "bearing lubrication deficiency" are both greater than the similarity calculation threshold, and the two values ​​are very close. At this point, bearing inner ring wear will not be output as the final fault diagnosis result alone. We can also re-analyze 0.984 / (0.984+0.981)≈50.1%, indicating that the two are equally likely. Therefore, we can generate candidate fault diagnosis results for the range hood based on bearing lubrication deficiency corresponding to 0.981. The interpretable output (which can be displayed in the fault diagnosis results) can be: suspected bearing-related problems, possibly inner ring wear or lubrication deficiency; it is recommended to focus on checking the bearing condition. Furthermore, we can also combine other related information in the preset knowledge graph (such as equipment runtime: if it is a new device, lubrication deficiency is more likely; if it is an old device, wear is more likely) to make a final judgment, so as to output more accurate fault diagnosis results and fault diagnosis reports.

[0117] Figure 6 This is a flowchart illustrating another method for diagnosing a range hood fault provided in an embodiment of the present invention, as shown below. Figure 6As shown, the multimodal signals of a range hood can include, but are not limited to, acoustic signals, vibration signals, and operating condition signals. At the same acquisition time, a built-in microphone array is used to acquire the acoustic signals of the range hood during operation, a vibration sensor is used to acquire the vibration signals, and a speed sensor is used to acquire the motor speed signals. These motor speed signals are then used as the operating condition signals of the range hood. Subsequently, feature extraction and compression are performed on the acoustic signals, vibration signals, operating condition signals, and their corresponding acquisition times to obtain feature vector data packets. These feature vector data packets can be transmitted to the cloud via a wireless communication module. The cloud access layer can receive concurrent data connections from millions of devices and perform efficient scheduling and preprocessing. The signal reconstruction and alignment layer can reassemble the received acoustic feature time sequence, vibration feature time sequence, and operating condition feature time sequence to reconstruct the specific scenario in which the data was generated; that is, to generate the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the operating condition feature time sequence corresponding to the operating condition signal. The data is then passed to the multimodal fusion diagnostic engine. The acoustic analysis sub-network learns patterns from the acoustic feature time-series, identifying abnormal sound patterns such as howling, friction, and imbalance. The vibration analysis sub-network learns patterns from the vibration feature time-series, identifying abnormal vibration patterns such as impact, resonance, and eccentricity. The attention fusion module synthesizes and summarizes the information to determine the fusion feature vector, facilitating subsequent diagnostic processes and model updates. On one hand, the final determined and output fault diagnosis results can include fault type, location, and confidence level (or matching value, similarity calculation value), and can also include maintenance suggestions for subsequent user reminders and after-sales systems.

[0118] Figure 7 This is a schematic diagram of a fault diagnosis device for a range hood provided in an embodiment of the present invention. This fault diagnosis device is applicable to detecting abnormal states during the operation of a range hood. The fault diagnosis device can be implemented in hardware and / or software and is generally configured in the control board. Figure 7 As shown, the fault diagnosis device includes:

[0119] The data acquisition and processing module 610 is used to synchronously acquire acoustic signals, vibration signals, and operating condition signals during the operation of the range hood, and perform data processing to obtain the acoustic feature time series corresponding to the acoustic signals, the vibration feature time series corresponding to the vibration signals, and the operating condition feature time series corresponding to the operating condition signals, respectively. The data fusion module 620 is used to perform data fusion processing on the acoustic feature time series, vibration feature time series, and operating condition feature time series using a multimodal fusion deep learning model based on an attention mechanism to obtain a fused feature vector. The similarity calculation module 630 is used to perform similarity matching analysis between the fused feature vector and multiple standard feature vectors in a preset knowledge graph to obtain multiple similarity calculation values. The fault diagnosis module 640 is used to determine the fault diagnosis result of the range hood based on the multiple similarity calculation values.

[0120] The technical solution in this invention collects acoustic signals, vibration signals, and operating condition signals during the operation of the range hood. Acquiring data from multiple information dimensions ensures further detailed analysis for fault diagnosis. A multimodal fusion deep learning model based on an attention mechanism is employed to fuse the acoustic feature time-series sequences corresponding to the acoustic signals, the vibration feature time-series sequences corresponding to the vibration signals, and the operating condition feature time-series sequences corresponding to the operating condition signals. This facilitates subsequent fault diagnosis of the range hood based on similarity matching analysis of the fused feature vectors. In other words, anomaly detection of the range hood can be performed using an acoustic-vibration fusion method based on multimodal fusion analysis of acoustic and vibration signals. Furthermore, remote fault diagnosis of the range hood can be performed using a pre-set knowledge graph in the cloud. This avoids the problem of cross-validation of historical big data and collaborative information caused by single-machine computation, effectively improving the monitoring accuracy of the range hood's operating condition and enhancing the judgment and diagnostic accuracy of the range hood system anomalies. The system can evolve with the accumulation of data from multiple fault diagnosis results, making the content of the pre-set knowledge graph in the cloud more consistent with the anomaly detection process of the range hood, thus enhancing its anti-interference capability.

[0121] Based on the above technical solution, optionally, the data acquisition and processing module 610 may specifically include a data acquisition unit, a data compression unit, and a data reconstruction unit. The data acquisition unit is used to synchronously acquire acoustic signals, vibration signals, and operating condition signals during the operation of the range hood. The data compression unit is used to perform feature extraction and data compression processing on the acoustic signals, vibration signals, operating condition signals, and corresponding acquisition times to obtain feature vector data packets. The data reconstruction unit is used to perform data reconstruction processing on the feature vector data packets in the time dimension to obtain the acoustic feature time sequence corresponding to the acoustic signals, the vibration feature time sequence corresponding to the vibration signals, and the operating condition feature time sequence corresponding to the operating condition signals, respectively.

[0122] Optionally, the data acquisition unit may specifically include a data acquisition subunit, which is used to acquire acoustic signals using a microphone array, acquire vibration signals using a vibration sensor, and acquire operating condition signals using a speed sensor at the same acquisition time.

[0123] Optionally, the data fusion module 620 may specifically include a weight allocation unit and a weighted fusion unit. The weight allocation unit is used to determine the weights corresponding to the acoustic feature time series, vibration feature time series, and operating condition feature time series respectively according to the multimodal fusion deep learning model based on the attention mechanism. The weighted fusion unit is used to perform weighted fusion processing on the acoustic feature time series, vibration feature time series, and operating condition feature time series according to the weights corresponding to the acoustic feature time series, vibration feature time series, and operating condition feature time series to obtain a fused feature vector.

[0124] Optionally, the similarity calculation module 630 may specifically include a similarity calculation unit, which is used to perform similarity calculation processing between the fused feature vector and each standard feature vector according to the cosine similarity principle, and obtain multiple similarity calculation values ​​accordingly.

[0125] Optionally, the fault diagnosis module 640 may specifically include a first similarity analysis unit and a matching fault diagnosis unit. The first similarity analysis unit is used to determine the similarity calculation value that satisfies a first condition among multiple similarity calculation values. The first condition is that the similarity calculation value is greater than the similarity calculation threshold and is the maximum value among the similarity calculation values. The matching fault diagnosis unit is used to take the determined fault mode as the matching fault diagnosis result of the range hood based on the standard feature vector corresponding to the similarity calculation value that satisfies the first condition and the preset mapping relationship between the standard feature vector and the fault mode.

[0126] Optionally, the fault diagnosis module 640 further includes a second similarity analysis unit and a candidate fault diagnosis unit. The second similarity analysis unit is used to determine the similarity calculation value that satisfies a second condition among multiple similarity calculation values. The second condition is that the similarity calculation value is greater than the similarity calculation threshold and less than the maximum value among the similarity calculation values. The candidate fault diagnosis unit is used to take the determined fault mode as the candidate fault diagnosis result of the range hood based on the standard feature vector corresponding to the similarity calculation value that satisfies the second condition and the preset mapping relationship between the standard feature vector and the fault mode.

[0127] The fault diagnosis device for range hoods provided in this embodiment of the invention can execute the fault diagnosis method for range hoods provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0128] Figure 8 This is a schematic diagram of the structure of a terminal device provided in an embodiment of the present invention, such as... Figure 8 As shown, the terminal device provided in this embodiment of the invention includes: one or more processors 71 and a storage device 72; the processors 71 in the terminal device may be one or more. Figure 8 Taking a processor 71 as an example; storage device 72 is used to store one or more programs; when one or more programs are executed by one or more processors 71, the one or more processors 71 implement the fault diagnosis method for range hoods provided in any of the embodiments of the present invention.

[0129] The terminal device may also include an input device 73 and an output device 74.

[0130] The processor 71, storage device 72, input device 73, and output device 74 in this terminal device can be connected via a bus or other means. Figure 8 Taking the example of a connection between China and Israel via a bus.

[0131] The storage device 72 in the terminal device serves as a readable storage medium and can be used to store one or more programs. These programs can be software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the fault diagnosis method for a range hood provided in this embodiment of the invention (e.g., attached...). Figure 7 The fault diagnosis device for the range hood shown includes the following modules: a data acquisition and processing module 610, used to simultaneously acquire acoustic signals, vibration signals, and operating condition signals during the operation of the range hood, and perform data processing to obtain acoustic feature time-series sequences corresponding to the acoustic signals, vibration feature time-series sequences corresponding to the vibration signals, and operating condition feature time-series sequences corresponding to the operating condition signals; a data fusion module 620, used to perform data fusion processing on the acoustic feature time-series sequences, vibration feature time-series sequences, and operating condition feature time-series sequences using a multimodal fusion deep learning model based on an attention mechanism to obtain a fused feature vector; a similarity calculation module 630, used to perform similarity matching analysis between the fused feature vector and multiple standard feature vectors in a preset knowledge graph to obtain multiple similarity calculation values; and a fault diagnosis module 640, used to determine the fault diagnosis result of the range hood based on the multiple similarity calculation values. The processor 71 executes various functional applications and data processing of the terminal device by running software programs, instructions, and modules stored in the storage device 72, thereby realizing the fault diagnosis method for the range hood in the above method embodiment.

[0132] Storage device 72 may include a stored program area and a stored data area, wherein the stored program area may store the operating system and applications required for at least one function; the stored data area may store data created based on the use of the device, etc. Furthermore, storage device 72 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, storage device 72 may further include memory remotely located relative to processor 71, and this remote memory may be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0133] Input device 73 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the device. Output device 74 may include display devices such as a display screen.

[0134] Furthermore, when one or more programs included in the aforementioned device are executed by one or more processors 71, the programs perform the following operations: synchronously acquire acoustic signals, vibration signals, and operating condition signals during the operation of the range hood, and perform data processing to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the operating condition feature time sequence corresponding to the operating condition signal, respectively; employ a multimodal fusion deep learning model based on an attention mechanism to perform data fusion processing on the acoustic feature time sequence, the vibration feature time sequence, and the operating condition feature time sequence to obtain a fused feature vector; perform similarity matching analysis between the fused feature vector and multiple standard feature vectors in a preset knowledge graph to obtain multiple similarity calculation values; and determine the fault diagnosis result of the range hood based on the multiple similarity calculation values.

[0135] This invention also provides a readable storage medium storing a computer program thereon, which, when executed by a processor, performs a fault diagnosis method for a range hood provided in any one of the embodiments of this invention. The control method includes:

[0136] Acoustic, vibration, and operating condition signals of the range hood are simultaneously collected and processed to obtain acoustic feature time-series sequences, vibration feature time-series sequences, and operating condition feature time-series sequences, respectively. A multimodal fusion deep learning model based on an attention mechanism is used to fuse these sequences, resulting in a fused feature vector. The fused feature vector is then compared with multiple standard feature vectors in a pre-defined knowledge graph to obtain multiple similarity calculation values. Based on these similarity calculation values, the fault diagnosis result of the range hood is determined.

[0137] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a readable storage medium. A readable storage medium can take many forms, including but not limited to: electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable CD-ROMs, optical storage devices, magnetic storage devices, or any suitable combination thereof. A readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0138] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0139] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, radio frequency (RF), etc., or any suitable combination thereof.

[0140] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including but not limited to a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0141] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0142] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for diagnosing faults in a range hood, characterized in that, include: The acoustic signals, vibration signals, and operating condition signals of the range hood are collected synchronously during operation and processed to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the operating condition feature time sequence corresponding to the operating condition signal, respectively. A multimodal fusion deep learning model based on an attention mechanism is used to perform data fusion processing on the acoustic feature time sequence, the vibration feature time sequence, and the working condition feature time sequence to obtain a fused feature vector; The fused feature vector is compared with multiple standard feature vectors in a preset knowledge graph for similarity matching analysis, resulting in multiple similarity calculation values. The fault diagnosis result of the range hood is determined based on multiple similarity calculation values.

2. The fault diagnosis method according to claim 1, characterized in that, The acoustic signals, vibration signals, and operating condition signals of the range hood are simultaneously acquired during operation and processed to obtain the acoustic feature time sequence corresponding to the acoustic signals, the vibration feature time sequence corresponding to the vibration signals, and the operating condition feature time sequence corresponding to the operating condition signals, including: The acoustic signal, vibration signal and operating condition signal of the range hood are collected simultaneously during operation; Feature extraction and data compression are performed on the acoustic signal, the vibration signal, the operating condition signal, and the corresponding acquisition time to obtain a feature vector data packet; In the time dimension, the feature vector data packet is reconstructed to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal, and the working condition feature time sequence corresponding to the working condition signal.

3. The fault diagnosis method according to claim 2, characterized in that, The acoustic signal, vibration signal, and operating condition signal of the range hood are collected synchronously during operation, including: At the same acquisition time, the acoustic signal is acquired using a microphone array, the vibration signal is acquired using a vibration sensor, and the operating condition signal is acquired using a speed sensor.

4. The fault diagnosis method according to claim 1, characterized in that, A multimodal fusion deep learning model based on an attention mechanism is used to perform data fusion processing on the acoustic feature time series, the vibration feature time series, and the operating condition feature time series to obtain a fused feature vector, including: Based on the attention-based multimodal fusion deep learning model, the weights corresponding to the acoustic feature time series, the vibration feature time series, and the working condition feature time series are determined respectively. Based on the weights corresponding to the acoustic feature time sequence, the vibration feature time sequence, and the operating condition feature time sequence, a weighted fusion process is performed on the acoustic feature time sequence, the vibration feature time sequence, and the operating condition feature time sequence to obtain the fused feature vector.

5. The fault diagnosis method according to claim 1, characterized in that, The fused feature vector is compared with multiple standard feature vectors in a preset knowledge graph using similarity matching analysis, resulting in multiple similarity calculation values, including: Based on the cosine similarity principle, the fused feature vector is compared with each of the standard feature vectors to calculate the similarity, resulting in multiple similarity values.

6. The fault diagnosis method according to claim 1, characterized in that, Based on multiple similarity calculation values, the fault diagnosis result of the range hood is determined, including: Among the multiple similarity calculation values, the similarity calculation value that satisfies the first condition is determined; wherein, the first condition is that it is greater than the similarity calculation threshold and is the maximum value among the similarity calculation values; Based on the standard feature vectors involved in the calculation corresponding to the similarity calculation values ​​that satisfy the first condition, and the preset mapping relationship between the standard feature vectors and the fault modes, the determined fault modes are used as the matching fault diagnosis results of the range hood.

7. The fault diagnosis method according to claim 6, characterized in that, Determining the fault diagnosis result of the range hood based on multiple similarity calculation values ​​also includes: Among the multiple similarity calculation values, the similarity calculation value that satisfies the second condition is determined; wherein the second condition is that it is greater than the similarity calculation threshold and less than the maximum value among the similarity calculation values; Based on the standard feature vectors involved in the calculation corresponding to the similarity calculation values ​​that satisfy the second condition, and the preset mapping relationship between the standard feature vectors and the fault modes, the determined fault modes are used as the candidate fault diagnosis results of the range hood.

8. A fault diagnosis device for a range hood, characterized in that, include: The data acquisition and processing module is used to synchronously acquire acoustic signals, vibration signals and operating condition signals during the operation of the range hood, and perform data processing to obtain the acoustic feature time sequence corresponding to the acoustic signal, the vibration feature time sequence corresponding to the vibration signal and the operating condition feature time sequence corresponding to the operating condition signal, respectively. The data fusion module is used to perform data fusion processing on the acoustic feature time series, the vibration feature time series and the working condition feature time series using a multimodal fusion deep learning model based on an attention mechanism to obtain a fused feature vector; The similarity calculation module is used to perform similarity matching analysis between the fused feature vector and multiple standard feature vectors in the preset knowledge graph, and obtain multiple similarity calculation values ​​accordingly. The fault diagnosis module is used to determine the fault diagnosis result of the range hood based on multiple similarity calculation values.

9. A terminal device, characterized in that, include: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the fault diagnosis method for the range hood as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the fault diagnosis method for the range hood as described in any one of claims 1-7.

Images