Surgical robot quality control and fault prediction method based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence and data processing, in particular to a surgical robot quality control and fault prediction method based on artificial intelligence, and the method specifically comprises the following steps: collecting the data of a surgical robot through a sensor, carrying out the manual marking of the collected data, and forming a sample data set; performing multi-scale normalization processing on the sample data set; constructing a fault prediction model of the surgical robot by adopting a convolutional neural network, and inputting normalized data into the model for training to obtain a trained fault prediction model; preprocessing the collected new data, inputting the preprocessed data into the trained fault prediction model, and outputting a fault category prediction result corresponding to the input data by the fault prediction model; and constructing a three-dimensional quality control matrix, wherein the three-dimensional quality control matrix triggers a corresponding automatic calibration program based on a category prediction result output by the fault prediction model. According to the method, the fault features of the surgical robot can be accurately captured, and the accuracy of fault prediction and the generalization ability of the model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and data processing, and particularly to a method for quality control and fault prediction of a surgical robot based on artificial intelligence. Background Art

[0002] With the extensive development of modern minimally invasive surgery, surgical robots, as high-precision and highly flexible operation platforms, have been widely used in many fields such as neurosurgery and urology. Their complex joint structures and multi-sensor control systems endow them with high controllability and repeatability when performing precise tasks. However, due to their high system integration and fast operation rhythm, any minor anomalies (such as mechanical jams, gear meshing errors, sensor drifts) may have an unignorable impact on surgical safety. Especially during the surgical process, some faults manifest as instantaneous shocks, frequency drifts or hysteresis phenomena, which are characterized by suddenness, locality and non-stationarity. Traditional quality control and fault detection methods are difficult to capture and accurately identify such complex signal features in a timely manner.

[0003] The Chinese invention patent with the publication number CN118938882A proposes an intelligent fault monitoring and diagnosis system and method for an industrial robot, which deploys target sensors at the monitoring parts of the target industrial robot to obtain real-time data; determines and analyzes the diagnosed faults based on machine learning technology according to the real-time data, and outputs a diagnostic report, which can specifically determine the monitoring parts to collect fault detection data and improve the efficiency of subsequent fault detection.

[0004] The Chinese invention patent with the publication number CN114897102A proposes a method, system, device and storage medium for fault diagnosis of an industrial robot, which mainly uses a twin neural network structure to realize automatic feature extraction and fault identification.

[0005] The Chinese invention patent with the publication number CN115099268A proposes an intelligent fault diagnosis method and system for a wheeled robot based on a graph convolutional network. The spatio-temporal difference graph convolutional network is used to calculate multi-order backward difference features of the data relationship graph of the wheeled robot, the local difference characteristics are used to enhance the features of the nodes, and the spatio-temporal graph convolutional module is used to obtain spatio-temporal related features. The constructed data relationship graph of the robot is beneficial to fault classification, and the developed STDGCN has the most advanced performance.

[0006] The objective drawbacks of the prior art are as follows: Conventional time-domain normalization methods fail to consider the frequency-band energy differences, amplify high-frequency noise, compress low-frequency features, and are prone to causing diagnostic failures; Conventional convolutional networks can only extract steady-state trends and cannot effectively capture transient shock features, making it easy to misjudge serious faults; The fixed-structure attention mechanism has serious redundant calculations and cannot focus on the fault-active period, resulting in low training efficiency; Fixed-bandwidth convolutions / filters cannot adapt to faults with frequency drift and lack frequency-band adaptability; Using cross-entropy loss ignores the imbalance of fault categories and the misjudgment cost, and the model tends to identify the majority class, losing its practicality; High-dimensional non-stationary signals lead to non-convergence or gradient explosion during training, with poor generalization ability.

[0007] Therefore, the present invention proposes an artificial-intelligence-based method for quality control and fault prediction of surgical robots to solve the above problems. Summary of the Invention

[0008] In view of the deficiencies of the prior art, the present invention develops an artificial-intelligence-based method for quality control and fault prediction of surgical robots. Through a neural network based on dual-path feature separation and enhancement, the present invention can accurately capture the fault features of surgical robots, improving the accuracy of fault prediction and the generalization ability of the model.

[0009] The technical solution for the present invention to solve the technical problem is an artificial-intelligence-based method for quality control and fault prediction of surgical robots, including the following steps: S1. Data acquisition: Through sensors installed on the surgical robot, collect the vibration signal data and operation feedback data of the surgical robot. At the same time, the edge acquisition device synchronously records the operation state of the surgical robot, and manually annotates the collected data in combination with the surgical log and expert advice. Then, unify the collected data into time-window sample data of a fixed length, and finally form a labeled sample data set. S2. Data preprocessing: Perform multi-scale normalization processing on the data in the sample data set. Specifically, adopt a normalization method based on frequency-band energy distribution, and at the same time combine wavelet decomposition to perform multi-scale energy equalization on the data to obtain the sample data after normalization processing. S3. Construct and train a fault prediction model: Use a convolutional neural network to construct a fault prediction model for the surgical robot, input the sample data after normalization processing into the fault prediction model for training, preset the number of iterative training times, and finally obtain a trained fault prediction model. The training process of the fault prediction model includes: constructing a convolutional neural network, performing feature extraction through a time series feature enhancement module, performing attention enhancement through a dynamic sparse self-attention mechanism, performing feature extraction through a residual adaptive frequency modulation module, performing feature fusion and alignment operations, performing multi-granularity time pooling operations, calculating a fault perception loss function, and updating the parameters of the neural network; S4. After preprocessing the newly collected data, input it into the trained fault prediction model of the surgical robot, and the fault prediction model outputs the fault category prediction result corresponding to the input data; S5. Construct a three-dimensional quality control matrix, including a real-time decision monitoring dimension, a maintenance decision dimension, and a process optimization dimension. The three-dimensional quality control matrix triggers the corresponding automatic calibration program based on the category prediction result output by the fault prediction model.

[0010] S1 is specifically as follows: The sensors installed on the surgical robot include a high-precision three-axis acceleration sensor and a force / torque sensor; The installation positions of the sensors are the key motion units, joint bearing parts, and execution ends of the surgical robot; The sensors continuously collect data at high frequency; The operating states of the surgical robot include but are not limited to joint angles, speeds, and control commands; Manual annotation means that experts mark faults according to abnormal working conditions. The abnormal working conditions include but are not limited to jamming, hysteresis, and abnormal vibration, and classify the fault marks, including but not limited to normal, slightly abnormal, and severely faulty; Slice the collected data, unify the collected data into time window samples of a fixed length, and at the same time label the corresponding fault category labels, and finally form a labeled sample data set.

[0011] S2 is specifically as follows: Normalize the data in the sample data set. The data in the sample data set is a vibration signal with characteristics of high-frequency noise, non-stationarity, and time-varying amplitude. Use a normalization method based on frequency band energy distribution, and at the same time combine wavelet decomposition to perform multi-scale energy equalization on the vibration signal. Specifically, decompose the input vibration signal into sub-band coefficients of different frequency bands through wavelet decomposition, calculate the energy mean and standard deviation of each layer of wavelet coefficients, normalize each layer of wavelet coefficients, and then dynamically adjust the normalized coefficient distribution through learnable scaling parameters and offset parameters, and finally obtain the normalized sample data.

[0012] Constructing a convolutional neural network is specifically as follows: The convolutional neural network structure specifically adopts a hierarchical architecture design. The input layer receives the vibration signals after multi-scale normalization and is sequentially connected to the temporal feature enhancement module, the dynamic sparse self-attention layer, the residual adaptive frequency modulation module, the feature fusion layer, and the fully connected layer; The temporal feature enhancement module separates and enhances the steady-state and transient features through a dual-path structure; the dynamic sparse self-attention layer uses a gating mechanism to focus on the fault active period; the residual adaptive frequency modulation module adaptively focuses on the frequency band through an adjustable filter; after the feature fusion layer integrates the input vibration signals and the filtered features, multi-granularity pooling is used to extract cross-scale statistical features, and finally the fault probability distribution is output through the fully connected layer.

[0013] The feature extraction through the temporal feature enhancement module is as follows: The temporal feature enhancement module is adopted to separate the steady-state and transient components of the signal through a dual-path structure, and then the long-term trend is extracted from the steady-state component through convolution operation, and the local mutations in the transient component are captured through wavelet transform; Specifically, the steady-state component is extracted through low-pass filtering convolution, the transient residual is obtained by subtracting the steady-state component from the input vibration signal, the steady-state component is enhanced by high-pass filtering, and at the same time the time-frequency features of the transient component are extracted by discrete wavelet transform. Finally, the features of the two paths are concatenated along the channel dimension.

[0014] The attention enhancement through the dynamic sparse self-attention mechanism is as follows: In the dynamic sparse self-attention mechanism module, a combined dynamic sparse gate is introduced. The attention score matrix is sorted row by row through the dynamic sparse gate, and only the Top-k significant associations are retained. Specifically, the first n maximum values in each row are retained, and the rest are set to zero, where n is initially 1 and linearly increases to the preset upper limit k with the number of training rounds, thereby gradually expanding the receptive field.

[0015] The feature extraction through the residual adaptive frequency modulation module is as follows: The residual adaptive frequency modulation module is used to embed an adjustable bandpass filter in the residual connection, enabling the network to adaptively focus on the fault-sensitive frequency band. The basic frequency band features are extracted through the convolution of the residual path, and a multi-layer perceptron is used to dynamically generate the band modulation factor to adaptively adjust the center frequency of the bandpass filter and focus on the fault-related frequency band.

[0016] The feature fusion and alignment operation and the multi-granularity time pooling operation are as follows: Feature fusion and alignment operation: The bandpass filtered features and the input vibration signals are fused, and normalized into a sequence of the same length as the input vibration signals. Specifically, a learnable gated convolutional upsampling module is used to expand the time dimension of the bandpass filtered features to T and perform gated feature fusion with the original vibration signals, dynamically allocating the residual weights to align the time axes; Multi - granularity time pooling operation: Adopt multi - scale time pooling operation, use multi - scale sliding windows to extract local statistical features. Different window sizes capture short - term shock and long - term trend features. Specifically, three different lengths of sliding windows are preset, with a window step size of 1. When sliding point - by - point on the time axis, two statistical metrics, namely the maximum value and the standard deviation within the window, are maintained in real - time. After the scanning is completed, the statistical features obtained from the three window lengths are concatenated in the channel direction, and redundant scale information is automatically suppressed through a layer with learnable gating, thereby forming comprehensive multi - scale time features. This can avoid blurring the fault - occurrence time period during pooling.

[0017] The calculation of the fault - aware loss function is as follows: Adopt a fault - aware loss function, combine class - sensitive weights with the fault characteristics of the surgical robot, and perform weighted focusing and misclassification penalty on fault - class samples; specifically, through dynamic class - weight allocation and misclassification cost - sensitive modulation, higher loss weights are imposed on fault - class samples, and additional penalties are imposed on key misclassification combinations. The calculation formula of the fault - aware loss function is as follows: , , Among them, represents the fault - aware loss function; represents the total number of sample data; represents the total number of fault classes; represents the class weight; represents the true class of the nth sample; represents the penalty coefficient when the true class of the nth sample is misclassified as the cth class; is the misclassification penalty factor; represents the logarithmic function with base 10; represents the focus factor to suppress easy - to - classify samples; represents the predicted probability that the nth sample belongs to the cth class.

[0018] The parameter update of the neural network is as follows: Update the parameters of the fault prediction model through adaptive gradient clipping and the AdamW optimizer. The calculation formula is as follows: , , , , , Among them, represents the gradient of the th iteration; represents the gradient of the fault-aware loss function with respect to the model parameters; represents the parameters of the convolutional neural network; represents the gradient after clipping in the th iteration; is the threshold parameter for gradient clipping; represents the L2 norm; represents the first moment estimate of the th iteration; represents the first moment estimate of the th iteration; represents the second moment estimate of the th iteration; represents the second moment estimate of the th iteration; represents the exponential decay rate of the second moment estimate; represents the parameters of the convolutional neural network in the th iteration; represents the parameters of the convolutional neural network in the th iteration; represents the learning rate parameter;

[0019] The effects provided in the Summary of the Invention are only the effects of the embodiments, rather than all the effects of the invention. The above technical solutions have the following advantages or beneficial effects: The present invention provides an artificial intelligence-based method for quality control and fault prediction of surgical robots. By using wavelet decomposition combined with dynamic normalization of frequency band energy, it can effectively retain low-frequency features and suppress high-frequency noise. By splitting the signal into two components, namely the steady state and the transient state, and processing and fusing them separately, it can solve the problem that transient shocks are smoothed during convolutional processing. By adopting a dynamic Top-k gating strategy, it can prevent the problem of information loss in the initial stage of training, be able to expand the receptive field in the later stage, and adaptively extract local fault-related features. By embedding an adjustable filter in the residual path and dynamically adjusting the center frequency in combination with the fully connected layer, it can achieve continuous tracking of time-varying frequency band faults. By adopting multi-scale sliding window maximum and standard deviation pooling and combining learnable gating, it can automatically suppress redundant scales. By adopting a dual modulation mechanism of class weight and misclassification cost, it can assign higher weights to specific fault class samples. By combining adaptive gradient clipping with the AdamW optimizer, it can improve the training stability and model generalization ability.

[0020] In summary, the present invention proposes a quality control and fault prediction method for surgical robots based on artificial intelligence. Through a neural network based on dual-path feature separation and enhancement, the problems of high-frequency noise interference, loss of transient shock features, difficulty in extracting local fault features, and time-variability of fault frequency bands in surgical robot fault detection are solved. It can accurately capture the fault features of surgical robots and improve the accuracy of fault prediction and the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention.

[0022] Figure 1 It is a schematic flowchart of the method of the present invention.

[0023] Figure 2 It is a schematic diagram of the input vibration signal.

[0024] Figure 3 It is a comparison chart of the time-frequency distribution of wavelet coefficients between the present invention and the conventional method.

[0025] Figure 4 It is a three-dimensional spectral waterfall chart of the minimum-maximum normalization method.

[0026] Figure 5 It is a three-dimensional spectral waterfall chart of the Z-score normalization method.

[0027] Figure 6 It is a three-dimensional spectral waterfall chart of the normalization method of the present invention.

[0028] Figure 7 It is an experimental chart of the dynamic sparse attention matrix.

[0029] Figure 8 It is a smooth curve chart of the attention sparsity.

[0030] Figure 9 It is a trend chart of the effective feature energy.

[0031] Figure 10 It is an experimental chart of the frequency response of the adaptive filter bank. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] In order to clearly illustrate the technical features of the present solution, the present invention will be described in detail below through specific embodiments and in combination with its drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the present invention. In order to simplify the disclosure of the present invention, the components and settings of specific examples are described below.

[0033] Embodiment 1 A quality control and fault prediction method for a surgical robot based on artificial intelligence, comprising the following steps: S1. Data acquisition: Vibration signal data and operation feedback data of the surgical robot are collected through sensors installed on the surgical robot. Meanwhile, the operation state of the surgical robot is synchronously recorded by edge acquisition devices, and the collected data is manually labeled in combination with surgical logs and expert suggestions. Then, the collected data is unified into time window sample data of a fixed length, and finally a labeled sample data set is formed; S2. Data preprocessing: The data in the sample data set is subjected to multi-scale normalization processing. Specifically, a normalization method based on frequency band energy distribution is adopted, and at the same time, wavelet decomposition is combined to perform multi-scale energy equalization on the data to obtain the sample data after normalization processing; S3. Constructing and training a fault prediction model: A convolutional neural network is used to construct a fault prediction model for the surgical robot. The sample data after normalization processing is input into the fault prediction model for training, and the number of iterative training times is preset in advance. Finally, a trained fault prediction model is obtained; The training process of the fault prediction model includes: constructing a convolutional neural network, performing feature extraction through a temporal feature enhancement module, performing attention enhancement through a dynamic sparse self-attention mechanism, performing feature extraction through a residual adaptive frequency modulation module, performing feature fusion and alignment operations, performing multi-granularity time pooling operations, calculating a fault perception loss function, and updating the parameters of the neural network; S4. The newly collected data is preprocessed and then input into the trained fault prediction model of the surgical robot, and the fault prediction model outputs the fault category prediction result corresponding to the input data; S5. Construct a three-dimensional quality control matrix, including a real-time decision monitoring dimension, a maintenance decision dimension, and a process optimization dimension. The three-dimensional quality control matrix triggers corresponding automatic calibration programs based on the category prediction results output by the fault prediction model.

[0034] In the specific implementation manner, S1 is as follows: The sensors installed on the surgical robot include high-precision three-axis acceleration sensors and force / torque sensors; The positions where the sensors are installed are the key motion units, joint bearing parts, and execution ends of the surgical robot; The sensors continuously collect data at high frequency; The operation states of the surgical robot include but are not limited to joint angles, speeds, and control commands; Manual labeling is that experts mark faults according to abnormal working conditions. Abnormal working conditions include but are not limited to jamming, hysteresis, and abnormal vibration, and the fault marks are classified, including but not limited to normal, slightly abnormal, and severely faulty; Slice the collected data to unify the collected data into time window samples of a fixed length, and at the same time label the corresponding fault category labels, finally forming a labeled sample data set.

[0035] In the specific implementation manner, S2 is as follows: Normalize the data in the sample data set. The data in the sample data set is a vibration signal with characteristics of high-frequency noise, non-stationarity, and time-varying amplitude. A normalization method based on frequency band energy distribution is adopted, and wavelet decomposition is combined to perform multi-scale energy equalization on the vibration signal. Specifically, the input vibration signal is decomposed into sub-band coefficients of different frequency bands through wavelet decomposition, the energy mean and standard deviation of each layer of wavelet coefficients are calculated, after normalizing each layer of wavelet coefficients, the normalized coefficient distribution is dynamically adjusted through learnable scaling parameters and offset parameters, and finally the normalized sample data is obtained.

[0036] The calculation formula for normalizing each layer of wavelet coefficients is as follows: , where, represents the normalized output of the wavelet decomposition coefficients of the th layer, representing the vibration energy of a specific frequency band; represents the th layer coefficient after the original vibration signal is decomposed by wavelet; represents the th layer coefficient energy mean; represents the th layer coefficient energy standard deviation; represents the trainable scaling parameter of the th layer coefficient, represents the trainable offset parameter of the th layer coefficient. The training method is gradient descent combined with the backpropagation algorithm, and and are updated through the chain rule to make them adapt to different frequency band feature distributions; represents the minimum constant to prevent division by zero; The energy mean is calculated instead of the amplitude mean because calculating the energy mean is more suitable for fault impact identification. The calculation formula for the energy mean is as follows: , where, represents the time series length; represents the th layer wavelet decomposition coefficient at time point ; By calculating the standard deviation of energy, the instability of the vibration amplitude can be better captured. The calculation formula for the energy standard deviation is as follows: ; It should be noted that by selecting the calculation based on the mean and standard deviation of the frequency band energy , rather than the traditional normalized time-domain statistic, it can dynamically adjust the normalization coefficient according to the frequency band energy for the non-stationarity of the vibration signal of the surgical robot, suppress high-frequency noise and retain low-frequency fault characteristics.

[0037] In the specific implementation manner, the convolutional neural network is constructed as follows: The convolutional neural network structure specifically adopts a hierarchical architecture design. The input layer receives the vibration signal after multi-scale normalization, and is sequentially connected to the time series feature enhancement module, the dynamic sparse self-attention layer, the residual adaptive frequency modulation module, the feature fusion layer, and the fully connected layer; The time series feature enhancement module separates and enhances the steady-state and transient features through a dual-path structure; the dynamic sparse self-attention layer uses a gating mechanism to focus on the active fault period; the residual adaptive frequency modulation module performs frequency band adaptive focusing through an adjustable filter; after the feature fusion layer integrates the input vibration signal and the filtered features, it extracts cross-scale statistical features through multi-granularity pooling, and finally outputs the fault probability distribution through the fully connected layer.

[0038] It should be noted that several (such as 3 layers) fully connected layers can be added to the middle layer of the convolutional neural network designed by the present invention, or the modules / layers proposed by the present invention can be deleted, which still belongs to the protection scope of the present invention.

[0039] In the specific implementation manner, the feature extraction through the time series feature enhancement module is as follows: The time series feature enhancement module is adopted to separate the steady-state and transient components of the signal through a dual-path structure, and then the long-term trend is extracted from the steady-state component through convolution operation, and the local mutation in the transient component is captured through wavelet transform. Mechanical jamming vibration signals include abnormalities such as instantaneous collisions and tool tip hysteresis in surgical robots, which are only visible in local windows. Using the time series feature enhancement module for feature extraction can solve the problem of transient feature loss and enhance the pertinence of the transient impact features of mechanical jamming vibration signals; Specifically, the steady-state component is extracted through low-pass filtering convolution, the input vibration signal is subtracted from the steady-state component to obtain the transient residual, and the steady-state component is enhanced through high-pass filtering. At the same time, the time-frequency features of the transient component are extracted using discrete wavelet transform, and finally the two-path features are concatenated along the channel dimension.

[0040] The calculation formula of the time series feature enhancement module is as follows: , , , where It represents a low-pass filter with a size of 5 and a cut-off frequency of 50 Hz. The steady-state components of the vibration signals of the surgical robot usually concentrate in the low frequency, usually with a frequency lower than 50 Hz, while transient impacts such as mechanical jamming are manifested as high-frequency mutations. It represents the vibration signal after multi-scale normalization. It represents the steady-state component, which is the signal after low-pass filtering. It characterizes the one-dimensional convolution with a step size of 1 and uses a high-pass filter. For Extract the low-frequency steady-state component. It represents the one-dimensional convolution operation. It is the residual between the input vibration signal and the steady-state component, characterizing the transient impact component and emphasizing transient anomalies such as mechanical jamming. It represents a high-pass filter. By enhancing the high-frequency details of the steady-state component through the high-pass filter, it can avoid the long-term trend from masking the transient features. It characterizes the one-dimensional convolution with a step size of 3 and uses a low-pass filter. For Extract the low-frequency steady-state component. It represents feature splicing. It represents the enhanced time-series features, fusing the time-frequency information of the steady-state and transient components. It represents the discrete wavelet transform, extracting the time-frequency features of the transient component and separating and locating the time-frequency mutation points of the transient.

[0041] In the specific implementation manner, the attention enhancement is carried out through the dynamic sparse self-attention mechanism as follows: In the dynamic sparse self-attention mechanism module, dynamic sparse gating is introduced. The attention score matrix is sorted row by row through dynamic sparse gating, and only the Top-k significant associations are retained. Specifically, the first n maximum values in each row are retained, and the rest are set to zero, where n is initially 1 and linearly grows to the preset upper limit k with the number of training rounds, thereby gradually expanding the receptive field and avoiding the loss of key features due to excessive sparsity in the early training.

[0042] The related calculations of the dynamic sparse self-attention mechanism are as follows: , , , , , , Among them, Represents a query matrix, characterizing the dynamic association pattern between time points, learning how to extract query patterns from input features, characterizing which feature dimensions need to be focused on at different time points, and optimized through the backpropagation algorithm; Represents a key matrix, characterizing the feature encoding pattern of time points, learning how to encode key patterns, characterizing how the features of time points are retrieved by other positions, and optimized through the backpropagation algorithm; Represents a value matrix, learning how to extract value patterns from input features, that is, the semantic representation of fault features, and optimized through the backpropagation algorithm; Represents a query vector matrix; Represents a key vector matrix; Represents the query vector of the th time point; Represents the key vector of the th time point; Represents the transpose of Represents the dimension of the key vector, characterizing the dimension scaling factor; Represents the attention score between the th time point and the th time point; Represents the element in the th row and th column of the sparse gating mask; Represents taking the top maximum values after sorting the attention scores of the th row, taking the top maximum values for each row. It should be noted that n is dynamically adjusted during training. Initially, n = 1 and gradually expands to a preset maximum threshold, such as 5, to avoid losing key features due to excessive sparsity in early training and gradually expand the receptive field to capture long-range dependencies; Represents the Softmax function; Represents the sparsified attention score matrix, whose elements are ; Represents the value vector matrix;

[0043] In the specific implementation manner, the feature extraction is specifically as follows through the residual adaptive frequency modulation module: Since the wear of surgical robot components and gear meshing errors often cause the fault frequency band to change at any time, the present invention embeds an adjustable band filter in the residual connection using the residual adaptive frequency modulation module, enabling the network to adaptively focus on the fault-sensitive frequency band, extracting the basic frequency band features through the convolution of the residual path, and dynamically generating a frequency band modulation factor using a multi-layer perceptron to adaptively adjust the center frequency of the band-pass filter to focus on the fault-related frequency band.

[0044] The calculation formula of the residual adaptive frequency modulation module is as follows: , , , where, represents the one-dimensional convolution of the residual path, and a residual convolution kernel is used as the filter bank to extract the basic frequency band features; represents the residual convolution kernel with a size of 3; represents the residual feature; represents the reference frequency band range; represents using a multi-layer perceptron, with the input to generate the frequency band modulation factor; represents the multi-layer perceptron with a hidden layer dimension of 64; represents the Sigmoid activation function; represents the Butterworth band-pass filter. By dynamically adjusting the filtering range, it can avoid the inability of traditional fixed-bandwidth convolution to adapt to time-varying fault frequency bands, and jointly optimize frequency modulation and feature enhancement through the residual path; represents performing dynamic frequency band filtering on the residual feature, and the center frequency is determined by controlled; represents the center frequency, which is dynamically generated by the multi-layer perceptron according to the input features. For the time-variability of the fault-related frequency bands, such as the resonance frequency shift caused by bearing wear, it adaptively focuses on the sensitive frequency bands; represents the reference frequency band, such as 50 to 150 Hz; represents the frequency band filtered feature.

[0045] In the specific implementation manner, the feature fusion and alignment operation and the multi-granularity time pooling operation are as follows: Feature fusion and alignment operation: Fuse the frequency band filtered feature with the input vibration signal, and normalize it into a sequence of the same length as the input vibration signal. Specifically, a learnable gated convolutional upsampling module is used to expand the time dimension of the frequency band filtered feature to T, and perform gated feature fusion with the original vibration signal, dynamically allocating residual weights to align the time axis; The calculation formula of the feature fusion and alignment operation is as follows: , , , where, represents the one-dimensional transposed convolution operation, using the convolution kernel to perform upsampling, with a stride is 2, representing upsampling the band filtering feature to a length , retaining the time localization information of the fault feature, such as the exact moment when the fault occurs; represents the upsampling convolutional kernel with a size of 4 and a stride of 2; represents the upsampled feature; represents concatenating the upsampled feature and the original vibration signal in the channel dimension; represents the gating convolutional kernel with a size of 3; represents the concatenated feature performs a one-dimensional convolutional operation using the convolutional kernel to extract the gating weight; represents the aligned feature after gated feature fusion; represents the gating weight matrix generated by the Sigmoid function, which is used to dynamically adjust the fusion ratio of the upsampled feature and the original signal, and solve the problem of time axis alignment between the band filtering feature and the original signal, such as the timing synchronization requirement of the surgical robot's actions; represents feature concatenation; represents element-wise multiplication.

[0046] Multi-granularity time pooling operation: A multi-scale time pooling operation is adopted to extract local statistical features using multi-scale sliding windows. Different window sizes capture short-term shock and long-term trend features. Specifically, three different lengths of sliding windows are preset with a window stride of 1. When sliding point by point on the time axis, two statistical indicators, the maximum value and the standard deviation within the window, are maintained in real time. After the scanning is completed, the statistical features obtained from the three window lengths are concatenated in the channel direction, and redundant scale information is automatically suppressed through a layer with learnable gating, thus forming comprehensive multi-scale time features, which can avoid blurring the fault occurrence time period during pooling.

[0047] The calculation formula of the multi-granularity time pooling operation is as follows: , , where, represents the pooling result of the feature; represents calculating the maximum value within the window; represents calculating the standard deviation within the window; represents the subsequence of the feature of within the window; represents the window size; represents the concatenation result of the multi-scale pooling features; Characterize the pooling results of different window sizes concatenated along the feature dimension, with sizes of 5, 10, and 20 respectively. For surgical robot failures, they may manifest as instantaneous anomalies or long-term performance degradation. Instantaneous anomalies such as gear fractures, and long-term performance degradation such as motor wear. Multi-scale can capture short-term shocks and long-term trends.

[0048] In the specific implementation manner, the calculation of the fault perception loss function is as follows: Adopt the fault perception loss function, combine the class-sensitive weight and the fault characteristics of the surgical robot, and perform weighted focusing and misclassification penalty on the fault class samples; specifically, through dynamic class weight allocation and misclassification cost-sensitive modulation, apply a higher loss weight to the fault class samples, and impose an additional penalty on the key misclassification combinations. The calculation formula of the fault perception loss function is as follows: , , Among them, represents the fault perception loss function; represents the total number of sample data; represents the total number of fault categories; represents the class weight; represents the true class of the th sample; represents the penalty coefficient when the true class of the th sample is misclassified as the cth class; ; represents the logarithmic function with base 10; represents the focus factor to suppress easy-to-classify samples, set ; represents the predicted probability that the nth sample belongs to the cth class.

[0049] Calculate the class weight through the frequency of the class , the class weight is inversely proportional to the square root of the frequency, suppressing the dominance of high-frequency classes in the loss. Mainly for the scarcity of surgical robot fault samples, such as intraoperative emergency shutdown events, it can enhance the attention to small-sample classes. The calculation formula of the class weight is as follows: , Among them, represents the frequency of the th fault class appearing,

[0050] In the specific implementation manner, the parameter update of the neural network is as follows: The calculation formula for updating the parameters of the fault prediction model through adaptive gradient clipping and the AdamW optimizer is as follows: , , , , , where, represents the gradient of the -th iteration; represents the gradient of the fault-aware loss function with respect to the model parameters; represents the parameters of the convolutional neural network; represents the gradient after clipping in the -th iteration; is the threshold parameter for gradient clipping, set to to limit the gradient norm and prevent gradient explosion caused by abnormal samples in the surgical robot data; represents the L2 norm; represents the first moment estimate in the -th iteration, characterizing the exponentially weighted moving average of the current gradient and used to smooth the gradient fluctuations; represents the first moment estimate in the -th iteration, characterizing the smoothed gradient result of the previous iteration; is the exponential decay rate of the first moment estimate; represents the second moment estimate in the -th iteration, characterizing the exponentially weighted moving average of the square of the current gradient and used to adaptively adjust the learning rate; represents the second moment estimate in the -th iteration, characterizing the weighted average of the squared gradient in the previous iteration; represents the exponential decay rate of the second moment estimate, set to ; represents the parameters of the convolutional neural network in the -th iteration; represents the parameters of the convolutional neural network in the -th iteration; represents the learning rate parameter, set to ; represents the weight decay coefficient, set to .

[0051] Example 2 As shown in Figure 2 and Figure 3As described above, the wavelet scale map is used to compare the transient shock feature capture capabilities of the time series feature enhancement module and the traditional convolutional network. By analyzing the simulated signal containing transient pulses and observing the time-frequency distribution of the wavelet coefficients after being processed by different methods, the short-time shock component is clearly visible in the time-domain waveform of the input vibration signal. However, in the wavelet coefficient map of the traditional method, only a weak response is presented during this period, indicating that the conventional convolutional operation smooths the transient features. The dual-path structure adopted in the present invention shows a strong red response region during the shock occurrence period, with the coefficient amplitude significantly increased and the boundary being clear. At the same time, the coefficient amplitude in the background noise region is suppressed. The improvement in time-frequency localization ability verifies the design advantage of separating the steady-state and transient components. By combining low-pass filtering and residual analysis, the sensitivity of the network to transient fault features such as mechanical jamming is enhanced.

[0052] As Figure 4 , Figure 5 , Figure 6 As shown, the regulation effects of different normalization methods on the energy of the frequency bands of the input vibration signal are compared through the three-dimensional spectrum waterfall diagram. According to the energy distribution of the input vibration signal of the surgical robot in the time-frequency domain, the energy change trends of high-frequency noise and low-frequency fault features are mainly observed. The input vibration signal shows energy fluctuations in the main fault feature frequency bands, and at the same time, noise energy aggregation appears in the high-frequency region. After adopting the traditional normalization method, the high-frequency noise energy is further amplified, forming an obvious red energy band, while the low-frequency feature energy shows irregular attenuation. In contrast, the multi-scale normalization processing of the present invention shows a more balanced energy distribution in the three-dimensional time-frequency diagram, maintaining a stable energy density in the low-frequency feature region, while presenting a significant blue attenuation band in the high-frequency noise region. The energy regulation characteristics verify that the normalization strategy based on the dynamic adjustment of the frequency band energy can effectively suppress the amplitude difference of non-stationary signals, avoid high-frequency noise interference, and at the same time retain the integrity of key fault features.

[0053] As Figure 7 , Figure 8 , Figure 9 As shown, the working principle and its advantages of the dynamic sparse self-attention mechanism are analyzed through the heat map and the training curve. The visualization of the attention matrix shows that the traditional method forms a diffuse energy distribution within the entire sequence, with a large number of inefficient weak associations. The matrix generated by the present invention presents a diagonal banded focusing pattern, forming a highlighted region only during the active period of the fault features. During the training process, the effective feature energy curve of the traditional method shows violent fluctuations, and the sparsity reaches the extreme value prematurely, resulting in the loss of key features. The present invention dynamically adjusts the sparse gating threshold, enabling the attention sparsity to increase progressively with the training process. Finally, while retaining the key time associations, the effective feature energy is stably maintained at a high level. The adaptive focusing mechanism not only avoids waste of computing resources but also ensures accurate feature extraction during the fault-sensitive period.

[0054] As shown Figure 10 As shown, the dynamic tracking ability of the residual adaptive frequency modulation module is verified through the frequency response curve of the filter bank. The scenario where the fault characteristic frequency drifts over time is analyzed experimentally. By comparing the fixed-bandwidth filter with the dynamic filter bank generated by the present technology, the experimental results show that the passband position of the fixed filter is constant, and its gain drops rapidly when the fault frequency shifts. In contrast, the center frequency of the filter bank generated by adaptive modulation can migrate continuously, always maintaining high-gain coverage of the characteristic frequency band, and the passband width is stable without distortion. In terms of the out-of-band noise suppression ability, the filter of the present invention exhibits a steep attenuation characteristic in the non-sensitive frequency band. The dynamic frequency focusing characteristic proves that the residual structure embedded with adjustable filters can effectively cope with the time-varying working conditions of mechanical systems and improve the signal ratio of fault characteristics.

[0055] Although the specific embodiments of the invention have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the invention. Based on the technical solutions of the present invention, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present invention.

Claims

1. A surgical robot quality control and fault prediction method based on artificial intelligence, characterized in that: The following steps are involved: S1. Data collection: The vibration signal data and operation feedback data of the surgical robot are collected through the sensors installed on the surgical robot. At the same time, the edge acquisition device synchronously records the operation status of the surgical robot. The collected data is manually annotated in combination with the surgical log and expert advice. The collected data is then unified into sample data of a fixed-length time window, and finally a sample data set with labels is formed. S2. Data preprocessing: Perform multi-scale normalization on the data in the sample data set. Specifically, a normalization method based on frequency band energy distribution is adopted. At the same time, multi-scale energy balance is performed on the data in combination with wavelet decomposition to obtain normalized sample data. S3. Construct a fault prediction model and perform training: Use a convolutional neural network to construct a fault prediction model for the surgical robot, input the normalized sample data into the fault prediction model for training, pre-set the number of iterative training times, and finally obtain a trained fault prediction model; The training process of the fault prediction model includes: building a convolutional neural network, extracting features through a temporal feature enhancement module, enhancing attention through a dynamic sparse self-attention mechanism, extracting features through a residual adaptive frequency modulation module, performing feature fusion and alignment operations, performing multi-granularity time pooling operations, calculating fault perception loss functions, and updating neural network parameters; S4, inputting the collected new data into the trained fault prediction model of the surgical robot after preprocessing, and the fault prediction model outputs the fault category prediction result corresponding to the input data; S5. Construct a three-dimensional quality control matrix, including real-time decision monitoring dimension, maintenance decision dimension and process optimization dimension. The three-dimensional quality control matrix triggers the corresponding automatic calibration program based on the category prediction results output by the fault prediction model.

2. The method for quality control and fault prediction of a surgical robot based on artificial intelligence according to claim 1 is characterized in that: S1 is as follows: The sensors installed on the surgical robot include high-precision three-axis acceleration sensors and force / torque sensors; The sensors are installed at the key motion units, joint bearings and execution ends of the surgical robot; Sensors continuously collect data at high frequencies; The operating status of the surgical robot includes but is not limited to joint angles, speeds, and control commands; Manual labeling is the process where experts mark faults based on abnormal operating conditions, including but not limited to jamming, hysteresis, and abnormal vibration, and classify fault markings, including but not limited to normal, slight abnormality, and severe fault; The collected data is sliced ​​and unified into time window samples of fixed length, and the corresponding fault category labels are marked to finally form a sample data set with labels.

3. The method for quality control and fault prediction of a surgical robot based on artificial intelligence according to claim 2 is characterized in that: S2 is as follows: The data in the sample data set are normalized. The data in the sample data set are vibration signals with high-frequency noise, non-stationarity and time-varying amplitude characteristics. A normalization method based on frequency band energy distribution is adopted, and wavelet decomposition is combined to perform multi-scale energy balance on the vibration signal. Specifically, the input vibration signal is decomposed into sub-band coefficients of different frequency bands through wavelet decomposition, and the energy mean and standard deviation of each layer of wavelet coefficients are calculated. After each layer of wavelet coefficients is normalized, the normalized coefficient distribution is dynamically adjusted through learnable scaling parameters and offset parameters, and finally the normalized sample data is obtained.

4. The method for quality control and fault prediction of a surgical robot based on artificial intelligence according to claim 3 is characterized in that: The details of building a convolutional neural network are as follows: The convolutional neural network structure adopts a hierarchical architecture design. The input layer receives the multi-scale normalized vibration signal, and sequentially connects the temporal feature enhancement module, the dynamic sparse self-attention layer, the residual adaptive frequency modulation module, the feature fusion layer, and the fully connected layer. The timing feature enhancement module separates and enhances steady-state and transient features through a dual-path structure; The dynamic sparse self-attention layer uses a gating mechanism to focus on fault-active periods; the residual adaptive frequency modulation module uses an adjustable filter to perform frequency band adaptive focusing; The feature fusion layer integrates the input vibration signal and the filtering features, and then extracts cross-scale statistical features through multi-granularity pooling, and finally outputs the fault probability distribution through the fully connected layer.

5. The method for quality control and fault prediction of a surgical robot based on artificial intelligence according to claim 4 is characterized in that: The feature extraction through the time series feature enhancement module is as follows: The time series feature enhancement module is used to separate the steady-state and transient components of the signal through a dual-path structure. Then, the long-term trend is extracted from the steady-state component through a convolution operation, and the local mutations in the transient component are captured through wavelet transform. Specifically, the steady-state component is extracted through low-pass filtering and convolution, the input vibration signal is subtracted from the steady-state component to obtain the transient residual, and then the steady-state component is enhanced by high-pass filtering. At the same time, the discrete wavelet transform is used to extract the time-frequency characteristics of the transient component, and finally the two path features are spliced ​​along the channel dimension.

6. The method for quality control and fault prediction of a surgical robot based on artificial intelligence according to claim 5 is characterized in that: The details of attention enhancement through dynamic sparse self-attention mechanism are as follows: The dynamic sparse self-attention mechanism module is combined with dynamic sparse gating. The attention score matrix is ​​sorted by row through dynamic sparse gating, and only the Top-k significant associations are retained. Specifically, the first n maximum values ​​in each row are retained, and the rest are set to zero, where n is initially 1 and increases linearly with the training rounds to the preset upper limit k, thereby gradually expanding the receptive field.

7. The method for quality control and fault prediction of a surgical robot based on artificial intelligence according to claim 6 is characterized in that: The feature extraction through the residual adaptive frequency modulation module is as follows: The residual adaptive frequency modulation module is used to embed an adjustable frequency band filter in the residual connection, so that the network can adaptively focus on the fault-sensitive frequency band. The basic frequency band characteristics are extracted through the convolution of the residual path. The frequency band modulation factor is dynamically generated using the multi-layer perceptron, and the center frequency of the bandpass filter is adaptively adjusted to focus on the fault-related frequency band.

8. The method for quality control and fault prediction of a surgical robot based on artificial intelligence according to claim 7 is characterized in that: The feature fusion and alignment operations and multi-granularity time pooling operations are as follows: Feature fusion and alignment operation: The frequency band filter features are fused with the input vibration signal and normalized into a sequence of the same length as the input vibration signal. Specifically, a learnable gated convolution upsampling module is used to expand the time dimension of the frequency band filter features to T, and gated feature fusion is performed with the original vibration signal, and residual weights are dynamically allocated to align the time axis. Multi-granularity time pooling operation: Multi-scale time pooling operation is adopted, and multi-scale sliding windows are used to extract local statistical features. Different window sizes capture short-term impact and long-term trend features. Three sliding windows of different lengths are preset, and the window step size is 1. When sliding point by point on the time axis, the two statistical indicators of the maximum value and standard deviation in the window are maintained in real time. After the scan is completed, the statistical features obtained by the three window lengths are spliced ​​in the channel direction, and the redundant scale information is automatically suppressed through the layer with learnable gating, thereby forming a comprehensive multi-scale time feature. This can avoid blurring the time period when the fault occurs during pooling.

9. The method for quality control and fault prediction of a surgical robot based on artificial intelligence according to claim 8 is characterized in that: The fault perception loss function is calculated as follows: The fault-aware loss function is used, combined with the category-sensitive weights and the fault characteristics of the surgical robot, to weighted focus and misclassification penalties on faulty samples. Specifically, through dynamic category weight allocation and misclassification cost-sensitive modulation, higher loss weights are imposed on faulty samples, and additional penalties are imposed on key misclassification combinations. The calculation formula of the fault-aware loss function is as follows: ， ， in, represents the fault-aware loss function; Indicates the total number of sample data; Indicates the total number of fault categories; represents the category weight; Indicates The true category of samples; Indicates The penalty coefficient when the true category of a sample is misclassified as the cth category; is the misclassification penalty factor; represents the logarithmic function with base 10; It indicates that the focus factor suppresses easily separable samples; It represents the predicted probability that the nth sample belongs to the cth category.

10. The method for quality control and fault prediction of a surgical robot based on artificial intelligence according to claim 9 is characterized in that: The parameters of the neural network are updated as follows: The parameter calculation formula for updating the fault prediction model through adaptive gradient clipping and AdamW optimizer is as follows: ， ， ， ， ， in, Indicates The gradient of the iteration; represents the gradient of the fault-aware loss function with respect to the model parameters; Represents the parameters of the convolutional neural network; Indicates The gradient after the clipping of the iteration; is the threshold parameter for gradient clipping; represents the L2 norm; Indicates The first moment estimate of the iteration; Indicates The first-order moment estimate of the iteration; Indicates The second moment estimate of the iteration; Indicates The second moment estimate of the iteration; represents the exponential decay rate of the second-order moment estimate; Indicates The parameters of the convolutional neural network at the iteration; Indicates The parameters of the convolutional neural network at the iteration; Represents the learning rate parameter; Represents the weight decay coefficient.

Images