A rehabilitation device fault data monitoring method and system

Abstract

The application discloses a kind of rehabilitation equipment failure data monitoring method and system, specifically related to electric digital data processing technical field, including S1, synchronous acquisition equipment operation time series data and patient interaction time series data, S2, two kinds of data are extracted and adaptively fused with features, S3, based on fusion features parallel execution equipment health prediction and rehabilitation curative effect deviation prediction, and the relevance of both is analyzed through cross attention mechanism, S4, combined with individualized dynamic baseline model, comprehensive equipment failure risk value, curative effect deviation and correlation confidence degree generate graded monitoring feedback.The application realizes the leap from simple failure early warning to treatment effectiveness guarantee by constructing equipment-patient-therapy ternary closed-loop analysis framework, solves the problem that the monitoring dimension of prior art is single and decoupled from clinical efficacy, and has the outstanding effect of strong early warning foresight, low false alarm rate and supporting root cause analysis.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and more specifically, to a method and system for monitoring fault data of rehabilitation equipment. Background Technology

[0002] Rehabilitation medical equipment plays an increasingly important role in modern clinical and home rehabilitation. Its operational reliability is directly related to treatment safety and efficacy. Currently, fault monitoring of rehabilitation equipment mainly draws on predictive maintenance technology for general industrial equipment. It usually relies on the equipment's own operating parameters, such as motor current, voltage, temperature, and vibration signals, and uses fixed thresholds or machine learning models to detect anomalies and provide early warnings of faults.

[0003] However, rehabilitation equipment has significant human-computer interaction characteristics. Its workload and operating status are closely related to the user's physiological characteristics, movement intentions and rehabilitation stage. Existing technical solutions have obvious limitations: treating rehabilitation equipment as ordinary industrial equipment for monitoring ignores human-computer interaction characteristics and is disconnected from therapeutic goals, resulting in inaccurate early warning and inability to guarantee the effectiveness of treatment.

[0004] In view of this, this application provides a method and system for monitoring fault data of rehabilitation equipment. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for monitoring fault data of rehabilitation equipment to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring fault data of rehabilitation equipment, comprising the following steps:

[0007] S1. Real-time synchronous collection of equipment operation sequence data during the operation of rehabilitation equipment, as well as patient interaction sequence data when patients use rehabilitation equipment;

[0008] S2. Perform feature extraction and fusion on the device runtime sequence data and patient interaction sequence data to obtain multimodal fusion features;

[0009] S3. Based on multimodal fusion features, the equipment health status prediction and rehabilitation efficacy deviation prediction are performed in parallel to obtain the equipment failure risk value and rehabilitation efficacy deviation. The correlation between the equipment failure risk value and rehabilitation efficacy deviation is analyzed to generate the correlation confidence score.

[0010] S4. Based on the equipment failure risk value, the deviation of rehabilitation efficacy, and the correlation confidence level, generate and output monitoring feedback information related to rehabilitation equipment failure or rehabilitation efficacy risk.

[0011] Preferably, in step S2, feature extraction and fusion are performed on the device runtime sequence data and the patient interaction sequence data. Specifically, this includes: extracting the device feature vector from the device runtime sequence data and the patient feature vector from the patient interaction sequence data; using a dynamic time warping algorithm to align the device feature vector sequence and the patient feature vector sequence in time; inputting the aligned feature vectors into a feature fusion network; and performing adaptive weighted fusion based on the signal-to-noise ratio of the two data streams and the prediction task weights to output multimodal fusion features.

[0012] Preferably, in step S3, the correlation between equipment failure risk value and rehabilitation efficacy deviation is analyzed to generate a correlation confidence score. Specifically, this involves obtaining the intermediate layer feature representation of the equipment health prediction sub-model used to generate the equipment failure risk value and the intermediate layer feature representation of the efficacy deviation prediction sub-model used to generate the rehabilitation efficacy deviation. The two sets of intermediate layer feature representations are input into the cross-attention coupling analysis module. By calculating the attention weight of equipment state features on efficacy deviation features and the attention weight of efficacy deviation features on equipment state features, a correlation confidence score that characterizes the strength of the causal relationship between the two is obtained.

[0013] Preferably, before step S4, the method further includes: establishing and maintaining a personalized dynamic baseline model for the corresponding combination of patient and rehabilitation equipment;

[0014] In step S4, the key indicators, equipment failure risk values, and deviation of rehabilitation efficacy in the multimodal fusion features are compared with the dynamic thresholds under the current working conditions provided by the personalized dynamic baseline model, which serve as the basis for generating monitoring feedback information.

[0015] The personalized dynamic baseline model is initialized based on data collected from the corresponding patient during a historical successful treatment cycle and is updated online as the treatment progresses.

[0016] Preferably, in step S3, the method for obtaining the deviation of rehabilitation efficacy is as follows: the treatment prescription for the current rehabilitation training is obtained in advance, the treatment prescription includes the expected rehabilitation effect indicators, the multimodal fusion features and the treatment prescription are input into the efficacy deviation prediction sub-model, the model outputs the predicted value of the rehabilitation effect indicators that can actually be achieved in this training, and the degree of deviation between the predicted value and the expected rehabilitation effect indicators is calculated as the deviation of rehabilitation efficacy.

[0017] Preferably, in step S4, generating and outputting monitoring feedback information related to rehabilitation equipment malfunction or rehabilitation efficacy risk specifically includes:

[0018] If the equipment failure risk value exceeds the first threshold and the correlation confidence exceeds the second threshold, a first type of early warning information is generated. The first type of early warning information indicates that the equipment failure is at high risk and may directly affect the rehabilitation effect.

[0019] If the equipment failure risk value does not exceed the first threshold, but the deviation of the rehabilitation efficacy exceeds the third threshold, and the correlation confidence is within the preset range, then a second type of warning information is generated. The second type of warning information indicates that there is a risk to the rehabilitation efficacy and suggests checking the equipment status.

[0020] If the deviation of the rehabilitation effect exceeds the third threshold, but the correlation confidence is lower than the preset range, a third type of warning information is generated. The third type of warning information provides guidance and suggestions for the patient's operation or status.

[0021] The present invention also provides a system for implementing the above-described method for monitoring fault data of rehabilitation equipment, the system comprising:

[0022] The multi-source data acquisition module is used to synchronously acquire real-time device operation sequence data of rehabilitation equipment and patient interaction sequence data of patients.

[0023] The data fusion processing module, connected to the multi-source data acquisition module, is used to extract and fuse features from device runtime time-series data and patient interaction time-series data to obtain multimodal fusion features;

[0024] The intelligent analysis engine connects to the data fusion and processing module, and includes:

[0025] The equipment health prediction sub-model is used to predict equipment failure risk values ​​based on multimodal fusion features;

[0026] A sub-model for predicting deviation in treatment efficacy is used to predict the degree of deviation in rehabilitation efficacy based on multimodal fusion features.

[0027] The coupling analysis module is used to analyze the correlation between equipment failure risk values ​​and the deviation of rehabilitation efficacy, and generate correlation confidence scores.

[0028] The decision feedback module connects to the intelligent analysis engine and is used to generate and output monitoring feedback information based on equipment failure risk values, deviation of rehabilitation efficacy, and correlation confidence.

[0029] Preferably, the system further includes:

[0030] The personalized baseline management module is used to establish and maintain a personalized dynamic baseline model for the corresponding patient and the corresponding rehabilitation equipment combination, and to provide dynamic thresholds for the decision feedback module.

[0031] The decision feedback module compares key indicators, equipment failure risk values, and deviations in rehabilitation efficacy from the multimodal fusion features with dynamic thresholds from the personalized baseline management module.

[0032] Preferably, the multi-source data acquisition module includes:

[0033] The equipment operation sensor group is integrated into the rehabilitation equipment to collect at least one of motor current, torque, encoder signal, structural stress, temperature and noise to form equipment operation sequence data.

[0034] The patient interaction sensor group, integrated into rehabilitation equipment or patient wearable devices, is used to collect at least one of surface electromyography signals, pressure distribution signals, inertial measurement unit signals, and joint motion visual signals to form patient interaction time-series data.

[0035] Preferably, the monitoring feedback information output by the decision feedback module is simultaneously pushed to the therapist terminal, the patient terminal, and the equipment maintenance management platform.

[0036] The technical effects and advantages of this invention are as follows:

[0037] 1. By introducing the evaluation dimension of rehabilitation efficacy deviation and constructing a dual-path analysis architecture of equipment health prediction and efficacy deviation prediction, it is possible to proactively identify the risk of decreased rehabilitation effect caused by hidden degradation of equipment performance before the equipment hardware parameters exceed the limit. This elevates the monitoring target from simply ensuring that the equipment does not stop to ensuring the safety and effectiveness of treatment, solving the long-standing problem of the disconnect between the monitoring target of existing technology and the core clinical needs.

[0038] 2. By establishing a personalized dynamic baseline model based on patients' historical success data, this system can generate dynamically changing health thresholds for different patients and different stages of rehabilitation, overcoming the false alarm and false negative problems caused by individual patient differences in the traditional fixed threshold method. Combined with the fusion analysis of equipment data and patient data, the system's warning targeting and reliability are fundamentally improved.

[0039] 3. By employing a cross-attention mechanism to couple the analysis of equipment failure risk and efficacy deviation, the correlation strength between the two can be quantified. This allows for a clear distinction between equipment failure, equipment performance deviation, and improper patient operation when issuing warnings. It provides therapists with a clear and accurate basis for adjusting treatment plans and equipment maintenance personnel with a clear basis for locating faults, greatly improving the efficiency of operation and maintenance and treatment. Attached Figure Description

[0040] Figure 1 This is a diagram illustrating the overall method steps of the present invention.

[0041] Figure 2 This is a system diagram of the present invention.

[0042] The attached diagram is labeled as follows: 1. Multi-source data acquisition module; 2. Data fusion and processing module; 3. Intelligent analysis engine; 30. Equipment health prediction sub-model; 31. Efficacy deviation prediction sub-model; 4. Coupled analysis module; 5. Decision feedback module; 6. Personalized baseline management module. Detailed Implementation

[0043] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0044] Example 1

[0045] As attached Figure 1 As shown, this embodiment of the invention provides a method for monitoring fault data of rehabilitation equipment, including the following steps:

[0046] S1, Real-time Synchronous Data Acquisition

[0047] During rehabilitation training, the multi-source data acquisition module 1 synchronously collects the device's runtime sequence data in real time. Interaction time series data with patients ;

[0048] Among them, device runtime sequence data Including but not limited to:

[0049] Motor current Torque ;

[0050] Encoder signal (position) ,speed );

[0051] Structural stress ;

[0052] temperature ;

[0053] noise level .

[0054] Patient interaction time series data Including but not limited to:

[0055] Surface electromyography signals ;

[0056] Pressure distribution signal ;

[0057] Inertial measurement unit signal (accelerometer) angular velocity ,angle );

[0058] Visual signals of joint motion (joint coordinates) ).

[0059] S2, Feature Extraction and Fusion

[0060] The collected device runtime sequence data Interaction time series data with patients Feature extraction and fusion are performed, specifically including:

[0061] S2.1 Feature Extraction

[0062] From device runtime timing data Extracting device feature vectors , including time-domain features, frequency-domain features, and time-frequency-domain features, among which;

[0063] Time-domain characteristics: mean, variance, peak value, and root mean square, etc.

[0064] Frequency domain characteristics: Spectral characteristics, such as the dominant frequency and band energy, are obtained through Fast Fourier Transform.

[0065] Time-frequency domain characteristics: Wavelet coefficient energy, etc., are obtained through wavelet transform;

[0066] From patient interaction time series data Extracting patient feature vectors It also includes time domain, frequency domain, and time-frequency domain features.

[0067] S2.2, Time Alignment

[0068] Because there may be a time synchronization issue between device data and patient data, an algorithm based on dynamic time warping is used to process the device feature vector sequence. and patient feature vector sequence Perform time alignment;

[0069] Dynamic time warping (RTW) achieves non-linear alignment of the time axis by finding the optimal matching path between two sequences. Let the two sequences... and The lengths are respectively and Dynamic Time Warping Algorithm Distance Calculated using the following recursive formula:

[0070]

[0071] in, Let be the Euclidean distance between two feature points. and Representing sequences respectively and In the and The feature vectors at each time point are used to find the optimal path through backtracking, thus achieving time alignment between the two sequences.

[0072] S2.3, Adaptive Weighted Fusion

[0073] The aligned feature vectors are input into a feature fusion network, which is a lightweight neural network whose output is a multimodal fused feature. The feature fusion network performs adaptive weighted fusion based on the signal-to-noise ratio of the two types of data streams and the prediction task weights;

[0074] Specifically, the feature fusion network consists of two fully connected layers and one fusion layer, where the aligned device feature vector is . The patient feature vector is First, feature mapping is performed using fully connected layers:

[0075]

[0076]

[0077] in, , and , These represent the weights and biases of the fully connected layers for device characteristics and patient characteristics, respectively. For activation functions;

[0078] Then, adaptive weights are calculated based on the signal-to-noise ratio and the prediction task weights. and :

[0079]

[0080]

[0081] in, and These are the signal-to-noise ratios of the device data stream and the patient data stream, respectively. and The weights for the equipment health prediction task and the efficacy deviation prediction task are respectively, which can be obtained by training with historical data or set by experts.

[0082] Finally, the features are fused. The calculation is as follows:

[0083]

[0084] S3, Parallel Prediction and Coupling Analysis

[0085] Based on multimodal fusion features Parallel execution of device health status prediction and rehabilitation efficacy deviation prediction, and analysis of the correlation between the two, specifically:

[0086] S3.1 Equipment Health Prediction

[0087] Multimodal fusion features The input device health prediction sub-model 30 is a sequence prediction model based on a long short-term memory network, and the output device failure risk value is... The value ranges from [0,1], and the larger the value, the higher the risk of failure.

[0088] The model structure consists of two LSTM layers and one fully connected output layer. Let the hidden states of the LSTM layers be... ,but:

[0089]

[0090]

[0091] in, It is the Sigmoid activation function. and The weights and biases of the output layer;

[0092] S3.2, Deviation prediction of rehabilitation efficacy

[0093] First, obtain the current rehabilitation training treatment prescription. This includes expected rehabilitation outcome indicators, such as joint range of motion targets. Muscle strength target wait;

[0094] Multimodal fusion features Treatment prescription The efficacy deviation prediction sub-model 31 is input together. This model is also based on the LSTM structure and outputs the predicted value of the rehabilitation effect index that can actually be achieved in this training. ;

[0095] Then, calculate the deviation of rehabilitation efficacy. :

[0096]

[0097] in, This is the target value corresponding to the treatment prescription;

[0098] S3.3 Cross-Attention Coupling Analysis

[0099] Obtain the intermediate layer feature representation of the equipment health prediction sub-model 30 (e.g., the hidden state of the last LSTM layer) and the intermediate layer feature representation of the efficacy deviation prediction sub-model 31 ;

[0100] Will and The input to the cross-attention coupling analysis module 4 calculates the attention weights of device state features on efficacy deviation features and the attention weights of efficacy deviation features on device state features, and then comprehensively derives the association confidence score. This is used to quantify the strength of the causal association between equipment failure risk and deviation from rehabilitation efficacy.

[0101] The specific calculation process is as follows:

[0102] First, calculate the query vector. Key vector Sum value vector :

[0103] , ,

[0104] , ,

[0105] in, , , , , and This is a trainable weight matrix; the subscript d2e indicates the association direction from device to therapeutic effect, and the subscript e2d indicates the association direction from therapeutic effect to device;

[0106] Then, the attention weights are calculated:

[0107]

[0108]

[0109] in, is the dimension of the key vector, used to scale the dot product;

[0110] Next, the weighted feature representation is calculated:

[0111]

[0112]

[0113] Finally, the attention features from both directions are concatenated, and the association confidence is output through a fully connected layer. :

[0114]

[0115] in, This represents vector concatenation. For the Sigmoid function, and For weights and biases, The value range is [0,1]. The larger the value, the stronger the correlation between equipment failure and deviation from the therapeutic effect.

[0116] S4, Personalized Dynamic Baseline and Decision Feedback

[0117] S4.1 Personalized Dynamic Baseline Model

[0118] For each patient and corresponding combination of rehabilitation equipment, a personalized dynamic baseline model is established and maintained. This model is initialized based on data collected during the patient's historical successful treatment cycles and is updated online as the treatment progresses.

[0119] Historical successful treatment cycles refer to the cycles in which patients complete training on the device and achieve the expected rehabilitation effect, while the device has no fault records. Multimodal fusion features are extracted from these cycles. Key metrics (such as feature mean, variance, etc.) are used to construct a Gaussian mixture model as the baseline model;

[0120] Let the baseline model be ,in, For the model parameters (including the weights, mean, and covariance of each Gaussian component), after each training session, if the training is marked as successful (i.e., the deviation from the therapeutic effect is below the threshold and the equipment is fault-free), then the training session will be marked as successful. Key metrics are added to the historical dataset, and GMM parameters are updated;

[0121] Online updates employ an incremental learning algorithm, as detailed below:

[0122]

[0123] in, For learning rate, The log-likelihood gradient;

[0124] Based on the baseline model, the current... Dynamic thresholds for key indicators. For example, using a 95% confidence interval as the normal range:

[0125]

[0126] S4.2 Decision Feedback

[0127] Multimodal fusion features Key indicators and equipment failure risk values and deviation of rehabilitation efficacy Compare with the dynamic thresholds provided by the personalized dynamic baseline model, combined with the association confidence score. Generate monitoring feedback information;

[0128] The specific decision-making rules are as follows:

[0129] If the equipment failure risk value > and > The system generates a first-class warning message, indicating a high risk of equipment malfunction that may directly affect the rehabilitation outcome, and recommends immediate shutdown for inspection and repair. The first threshold is (e.g., 0.8). The second threshold (e.g., 0.7);

[0130] If the equipment failure risk value ≤ However, the deviation of rehabilitation efficacy > and ≤ ≤ This generates a second type of early warning information, indicating a risk to the rehabilitation efficacy and suggesting checking the equipment status (such as calibration parameters). The third threshold (e.g., 0.2). and The lower and upper limits of the preset range (e.g., 0.3 and 0.6);

[0131] If the rehabilitation effect deviates > but < The system generates a third type of warning message, providing guidance and suggestions for the patient's actions or condition, such as "please relax your muscles" or "please adjust your posture."

[0132] The generated monitoring feedback information is simultaneously pushed to the therapist terminal, patient terminal, and equipment maintenance management platform through the decision feedback module 5, so that all parties can take appropriate measures in a timely manner.

[0133] Example 2

[0134] As attached Figure 2 As shown, this embodiment of the invention also provides a system for implementing the rehabilitation equipment fault data monitoring method in Embodiment 1 above. The system specifically includes the following modules:

[0135] Multi-source data acquisition module 1: includes a device operation sensor group and a patient interaction sensor group, used to synchronously acquire device operation sequence data of rehabilitation equipment and patient interaction sequence data in real time;

[0136] Data fusion processing module 2: Connects to multi-source data acquisition module 1, and is used to extract and fuse features from device runtime time-series data and patient interaction time-series data to obtain multimodal fusion features;

[0137] Intelligent Analysis Engine 3: Connects to Data Fusion Processing Module 2, including:

[0138] Equipment health prediction sub-model 30: used to predict equipment failure risk values ​​based on multimodal fusion features;

[0139] Therapeutic efficacy deviation prediction sub-model 31: used to predict the deviation of rehabilitation efficacy based on multimodal fusion features;

[0140] Coupled analysis module 4: Used to analyze the correlation between equipment failure risk value and deviation of rehabilitation efficacy, and generate correlation confidence score;

[0141] Decision Feedback Module 5: Connects to Intelligent Analysis Engine 3, and is used to generate and output monitoring feedback information based on equipment failure risk value, deviation of rehabilitation efficacy and correlation confidence.

[0142] Personalized baseline management module 6: Used to establish and maintain a personalized dynamic baseline model for the corresponding patient and corresponding rehabilitation equipment combination, and to provide dynamic thresholds for decision feedback module 5.

[0143] The following is a specific example of upper limb rehabilitation training to illustrate the implementation process of this invention:

[0144] Data acquisition: Patients use the upper limb rehabilitation robot for flexion and extension training. The device's operating sensors collect data such as motor current and encoder position, while the patient interaction sensors collect electromyographic signals on the arm surface and joint angle data.

[0145] Feature fusion: Extract the frequency domain features (dominant frequency, harmonic energy) of the motor current and the time domain features (root mean square value) of the electromyography signal, align them using DTW, and then obtain the fused features through a feature fusion network;

[0146] Prediction and Analysis: The equipment health prediction sub-model 30 outputs a failure risk value of 0.15 (low risk). The efficacy deviation prediction sub-model 31, combined with the treatment prescription (target joint range of motion 60°), predicts the actual range of motion as 55°, calculates the efficacy deviation as 0.083, and calculates the association confidence as 0.25 using the cross-attention module.

[0147] Baseline Comparison and Decision: Query the patient's personalized dynamic baseline. The current fusion characteristics are within the normal range. Due to the low risk of failure, low deviation of efficacy, and low confidence of association, the system does not issue a fault warning, but generates a prompt message: The activity level of this training is slightly lower than the target. Please pay attention to completing the actions according to the standard.

[0148] As can be seen from the above embodiments, the present invention can comprehensively utilize equipment and patient data to achieve accurate fault monitoring and efficacy assurance.

[0149] In summary:

[0150] The working principle of this application is based on multimodal data fusion and intelligent correlation analysis: First, the operation sequence data of rehabilitation equipment (such as current and vibration) and the interaction sequence data of patients (such as electromyographic signals and movement angles) are collected simultaneously. Then, feature extraction and adaptive fusion are performed on the two types of data to form a unified multimodal feature representation. Next, the device health prediction and rehabilitation efficacy deviation prediction are performed in parallel using this feature. The causal correlation strength between the two prediction results is analyzed through a cross-attention mechanism to generate correlation confidence. Finally, combined with the adaptive threshold provided by the personalized dynamic baseline model, the device failure risk value, efficacy deviation, and correlation confidence are integrated to generate graded early warning and decision suggestions, thereby achieving a leap from simple device failure early warning to ensuring the effectiveness of human-machine system treatment.

[0151] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method of monitoring rehabilitation equipment failure data, characterized by, The method comprises the following steps: S1, real-time synchronous acquisition of equipment running time sequence data of the rehabilitation equipment in the running process and patient interaction time sequence data when the patient uses the rehabilitation equipment; S2, feature extraction and fusion of the equipment running time sequence data and the patient interaction time sequence data to obtain multi-modal fusion features; S3, based on the multi-modal fusion features, parallel execution of equipment health state prediction and rehabilitation efficacy deviation prediction to obtain equipment failure risk value and rehabilitation efficacy deviation degree, and analysis of the correlation between the equipment failure risk value and the rehabilitation efficacy deviation degree to generate correlation confidence; S4, based on the equipment failure risk value, the rehabilitation efficacy deviation degree and the correlation confidence, generating and outputting monitoring feedback information related to the rehabilitation equipment failure or rehabilitation efficacy risk; In step S3, the correlation between the equipment failure risk value and the rehabilitation efficacy deviation degree is analyzed to generate the correlation confidence. Specifically, the intermediate layer feature representation of the equipment health prediction sub-model (30) used to generate the equipment failure risk value and the intermediate layer feature representation of the efficacy deviation prediction sub-model (31) used to generate the rehabilitation efficacy deviation degree are obtained. The two sets of intermediate layer feature representations are input into the cross-attention coupling analysis module (4). The correlation confidence representing the causal correlation strength between the two is obtained by calculating the attention weight of the equipment state feature on the efficacy deviation feature and the attention weight of the efficacy deviation feature on the equipment state feature. The specific steps include: An intermediate layer feature representation of the device health prediction submodel (30) is obtained An intermediate layer feature representation of the and efficacy deviation prediction submodel (31) is obtained ; Will And The input cross-attention coupling analysis module (4) comprehensively obtains the correlation confidence by calculating the attention weight of the device state feature to the efficacy deviation feature and the attention weight of the efficacy deviation feature to the device state feature , used to quantify the causal correlation strength between the device failure risk and the rehabilitation efficacy deviation; The specific calculation process is as follows: First, a query vector is calculated , a key vector and a value vector : ， ， ， ， wherein, , , , , and are trainable weight matrices; subscript represents a direction of association of the device to the efficacy, represents a direction of association of the efficacy to the device; Then, the attention weight is calculated: wherein, is the dimension of the key vector used to scale the dot product; Next, the weighted feature representation is calculated: Finally, the attention features of two directions are concatenated and the correlation confidence is output through a fully connected layer : wherein, denotes vector concatenation, is a Sigmoid function, and are weights and bias, the value range of is [0, 1], the greater the value, the stronger the relevance of the device failure and the deviation of the efficacy.

2. The rehabilitation equipment failure data monitoring method of claim 1, wherein, In step S2, the equipment running time sequence data and the patient interaction time sequence data are subjected to feature extraction and fusion, which specifically includes: extracting the equipment feature vector of the equipment running time sequence data and the patient feature vector of the patient interaction time sequence data, performing time alignment on the equipment feature vector sequence and the patient feature vector sequence using an algorithm based on dynamic time warping, inputting the aligned feature vectors into a feature fusion network, and outputting multi-modal fusion features by the feature fusion network according to the signal-to-noise ratio of the two types of data streams and the prediction task weight.

3. The rehabilitation equipment failure data monitoring method of claim 2, wherein, Before step S4, a personalized dynamic baseline model for the corresponding patient and the corresponding rehabilitation equipment combination is established and maintained; In step S4, the key indicators in the multi-modal fusion features, the equipment failure risk value and the rehabilitation efficacy deviation degree are compared with the dynamic threshold provided by the personalized dynamic baseline model under the current working condition, which serves as the basis for generating the monitoring feedback information; The personalized dynamic baseline model is initialized according to the data collected during the historical successful treatment period of the corresponding patient and is updated online during the treatment process; The historical successful treatment cycle refers to a cycle in which a patient completes training on the device and achieves an expected rehabilitation effect, while the device has no fault record, and multi-modal fusion features are extracted from these cycles Key indicators, construct a Gaussian mixture model as a personalized dynamic baseline model; Let the personalized dynamic baseline model be where, is the model parameter, after each training, if this training is marked as successful, the current key indicators are added to the historical dataset, and the GMM parameters are updated; The online update adopts an incremental learning algorithm, which is as follows: wherein, is the learning rate, is the log-likelihood gradient; According to the baseline model, calculate the current Dynamic threshold for key indicators: 。 4. The rehabilitation equipment failure data monitoring method of claim 3, wherein, In step S3, the rehabilitation efficacy deviation degree is obtained by: pre-acquiring the treatment prescription of the current rehabilitation training, the treatment prescription containing the expected rehabilitation effect indicators, inputting the multi-modal fusion features and the treatment prescription into the efficacy deviation prediction sub-model (31), the model outputting the predicted value of the actual achievable rehabilitation effect indicators of this training, and calculating the deviation degree between the predicted value and the expected rehabilitation effect indicators as the rehabilitation efficacy deviation degree.

5. The rehabilitation equipment failure data monitoring method of claim 4, wherein, In step S4, monitoring feedback information related to the rehabilitation device failure or rehabilitation effectiveness risk is generated and output, specifically including: If the device failure risk value exceeds the first threshold value, and the associated confidence exceeds the second threshold value, a first type of warning information is generated, indicating a high risk of device failure and may directly affect the rehabilitation effectiveness; If the device failure risk value does not exceed the first threshold value, but the rehabilitation effectiveness deviation exceeds the third threshold value, and the associated confidence is within a preset range, a second type of warning information is generated, prompting the existence of rehabilitation effectiveness risk and suggesting to check the device status; If the rehabilitation effectiveness deviation exceeds the third threshold value, but the associated confidence is below the preset range, a third type of warning information is generated, providing guidance suggestions for patient operation or state.

6. A rehabilitation equipment failure data monitoring system for implementing the rehabilitation equipment failure data monitoring method of claim 5, characterized by, The system comprises: A multi-source data acquisition module (1) for real-time synchronous acquisition of device runtime data of the rehabilitation device and patient interaction time series data of the patient; A data fusion processing module (2) connected to the multi-source data acquisition module (1) for feature extraction and fusion of the device runtime data and the patient interaction time series data to obtain multi-modal fusion features; An intelligent analysis engine (3) connected to the data fusion processing module (2), comprising: A device health prediction sub-model (30) for predicting a device failure risk value based on the multi-modal fusion features; An effectiveness deviation prediction sub-model (31) for predicting a rehabilitation effectiveness deviation based on the multi-modal fusion features; A coupling analysis module (4) for analyzing the correlation between the device failure risk value and the rehabilitation effectiveness deviation to generate an associated confidence; A decision feedback module (5) connected to the intelligent analysis engine (3) for generating and outputting monitoring feedback information based on the device failure risk value, the rehabilitation effectiveness deviation, and the associated confidence; wherein the intermediate layer feature representation of the device health prediction submodel (30) and the intermediate layer feature representation of the efficacy deviation prediction submodel (31) ; Will And The input cross-attention coupling analysis module (4) comprehensively obtains the correlation confidence by calculating the attention weight of the device state feature to the treatment effect deviation feature and the attention weight of the treatment effect deviation feature to the device state feature , for quantifying the causal correlation strength between the device failure risk and the rehabilitation treatment effect deviation; The specific calculation process is as follows: First, the query vector is computed , the key vector and the value vector : ， ， ， ， wherein, , , , , and are trainable weight matrices; subscript represents a direction of association of the device to the efficacy, represents a direction of association of the efficacy to the device; Then, the attention weight is calculated: wherein, is the dimension of the key vector used to scale the dot product; Next, the weighted feature representation is calculated: Finally, the attention features of two directions are concatenated and the correlation confidence is output through a fully connected layer : wherein, denotes vector concatenation, is a Sigmoid function, and are weights and bias, The value range of is [0, 1], and the larger the value, the stronger the relevance of the device failure and the deviation of the efficacy.

7. The system of claim 6, wherein, The system further comprises: An individualized baseline management module (6) for establishing and maintaining an individualized dynamic baseline model for the corresponding patient and the corresponding rehabilitation device combination, and providing dynamic thresholds for the decision feedback module (5); The individualized dynamic baseline model is initialized according to the data collected during the historical successful treatment period of the corresponding patient, and is updated online as the treatment progresses; The historical successful treatment cycle refers to a cycle in which a patient completes training on the device and achieves an expected rehabilitation effect, while the device has no fault record, and multi-modal fusion features are extracted from these cycles The key indicators are used to construct a Gaussian mixture model as a personalized dynamic baseline model. Let the personalized dynamic baseline model be where, is the model parameter, after each training, if this training is marked as successful, the key indicators of this time are added to the historical data set, and the GMM parameters are updated; The online update adopts an incremental learning algorithm, specifically as follows: wherein, is the learning rate, is the log-likelihood gradient; According to the personalized dynamic baseline model, calculate the current Dynamic threshold of key indicators: ； The decision feedback module (5) compares the key indicators in the multi-modal fusion features, the device failure risk value, and the rehabilitation effectiveness deviation with the dynamic thresholds from the individualized baseline management module (6).

8. The system of claim 7, wherein, The multi-source data acquisition module (1) comprises: A device operation sensor group integrated in the rehabilitation device for collecting at least one of motor current, torque, encoder signal, structural stress, temperature, and noise to form device runtime data; A patient interaction sensor group integrated in the rehabilitation device or patient wearable device for collecting at least one of surface electromyography signal, pressure distribution signal, inertial measurement unit signal, and joint motion visual signal to form patient interaction time series data.

9. The system of claim 8, wherein, The monitoring feedback information output by the decision feedback module (5) is simultaneously pushed to a therapist terminal, a patient terminal and a device maintenance management platform.

Images