Integrated circuit fault detection method based on artificial intelligence

Abstract

The invention relates to the technical field of fault detection, in particular to an integrated circuit fault detection method based on artificial intelligence. Comprising the following steps: acquiring and preprocessing multi-modal signal data and multi-modal historical signal data of an integrated circuit, and performing feature extraction on the preprocessed multi-modal signal data and multi-modal historical signal data by using a multi-dimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm, multi-modal signal feature data and multi-modal historical signal feature data are obtained; constructing and training an intelligent fault detection model based on the multi-modal historical signal feature data; performing optimization processing on the intelligent fault detection model by using an intelligent fault detection optimization algorithm to obtain an optimized intelligent fault detection model; and processing the multi-modal signal characteristic data to obtain a fault detection result. The technical problem that in the fault detection process of the integrated circuit, signal data processing is not accurate enough, and consequently the precision is not high in the fault detection process is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and in particular to an integrated circuit fault detection method based on artificial intelligence. Background Art

[0002] In the field of integrated circuit fault detection, with the continuous improvement of the functions and complexity of electronic devices, the accuracy and efficiency of fault detection systems have become important technical challenges in the design and manufacturing process. Traditional fault detection methods usually rely on rule-based detection techniques or model-based statistical methods, which often rely on a deep understanding of the circuit and a large amount of prior knowledge. Therefore, when faced with complex fault modes, the applicability and accuracy of these methods are greatly limited. In addition, integrated circuits generate multimodal signals during operation, including voltage, current, and power, which reflect the different working states and dynamic changes inside the circuit. Traditional fault detection methods often cannot effectively integrate these multi-dimensional and multi-scale signal features, resulting in the inability to fully capture subtle changes in signals and potential fault information during signal analysis and feature extraction. Therefore, traditional fault detection methods have certain limitations in practical applications, especially when dealing with dynamic signals and complex fault modes, the accuracy and efficiency of fault diagnosis are difficult to guarantee.

[0003] However, the existing integrated circuit fault detection methods have the following technical problems: in integrated circuit fault detection, the signal data processing is not accurate enough, resulting in low accuracy in the fault detection process and lack of sufficient robustness. Summary of the invention

[0004] The present invention provides an integrated circuit fault detection method based on artificial intelligence to solve the technical problems in integrated circuit fault detection, that is, inaccurate processing of signal data therein leads to low accuracy in the fault detection process and lack of sufficient robustness.

[0005] An integrated circuit fault detection method based on artificial intelligence of the present invention specifically includes the following technical solutions:

[0006] An integrated circuit fault detection method based on artificial intelligence comprises the following steps:

[0007] S1. Acquire and preprocess the multimodal signal data and multimodal historical signal data of the integrated circuit to obtain the preprocessed multimodal signal data and multimodal historical signal data, and use the multidimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm to extract features from the preprocessed multimodal signal data and multimodal historical signal data to obtain multimodal signal feature data and multimodal historical signal feature data; construct and train an intelligent fault detection model based on the multimodal historical signal feature data to obtain a final intelligent fault detection model;

[0008] S2. Optimize the intelligent fault detection model using an intelligent fault detection optimization algorithm to obtain an optimized intelligent fault detection model; process the multimodal signal feature data using the optimized intelligent fault detection model to obtain the fault detection result of the integrated circuit.

[0009] Preferably, the S1 specifically includes:

[0010] The multi-dimensional time-domain-frequency-nonlinear feature fusion extraction algorithm extracts features from pre-processed multi-modal signal data and multi-modal historical signal data by introducing adaptive multi-component decomposition, nonlinear dynamics analysis and multi-scale dynamic characteristic fusion.

[0011] Preferably, the S1 specifically includes:

[0012] In the implementation process of the multidimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm, the decomposition basis is constructed by combining the time domain, frequency domain and nonlinear modulation characteristics; in the adaptive multi-component decomposition process, based on the role of the decomposition basis, the preprocessed multimodal signal data is decomposed to obtain sub-component signals.

[0013] Preferably, the S1 specifically includes:

[0014] In the implementation process of the multidimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm, the local nonlinear characteristics of each sub-component signal are analyzed to obtain the local deviation degree and nonlinear energy ratio of the sub-component signal, and the local deviation degree and nonlinear energy ratio of the sub-component signal are mapped into a nonlinear dynamic characteristic matrix.

[0015] Preferably, the S1 specifically includes:

[0016] In the implementation process of the multi-dimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm, multi-scale dynamic characteristics fusion processing is performed on each sub-component signal, and a Gaussian convolution kernel is introduced to extract multi-scale dynamic characteristics. The convolution results of the sub-component signal at different scales are obtained, and the energy of the convolution result is calculated as the weight of each scale, and finally the comprehensive dynamic characteristics of the sub-component signal are obtained.

[0017] Preferably, the S1 specifically includes:

[0018] In the implementation process of the multidimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm, a nonlinear interaction characteristic matrix is ​​defined to determine the nonlinear interaction between sub-component signals; and the nonlinear dynamic characteristic matrix, comprehensive dynamic characteristics and nonlinear interaction characteristics of the sub-component signals are combined to obtain a comprehensive characteristic matrix, that is, multimodal signal feature data.

[0019] Preferably, the S2 specifically includes:

[0020] In the intelligent fault detection optimization algorithm, the intelligent fault detection model is updated with gradients and the learning rate is adjusted. By introducing a dynamic learning rate adjustment mechanism, the learning rate is dynamically adjusted in each iteration in combination with the fluctuation of the current gradient. The specific implementation formula is:

[0021]

[0022] in, It is The learning rate in the iteration; η0 is the initial learning rate; λ1, λ2 are parameters that control the decay speed of the learning rate; q, is the time scale for controlling the attenuation of the learning rate; γ1 is the factor for dynamically adjusting the learning rate.

[0023] Preferably, the S2 specifically includes:

[0024] In the intelligent fault detection optimization algorithm, the gradient update process is corrected by combining nonlinear gradient adjustment with nonlinear activation function. A spatial weighting mechanism is introduced to perform weighted update on each multimodal historical signal feature data through spatial weighting factors.

[0025] Preferably, the S2 specifically includes:

[0026] In the intelligent fault detection optimization algorithm, the time-space cross-feedback mechanism is introduced to optimize the learning process of the intelligent fault detection model, and finally the optimized intelligent fault detection model is obtained. The expression of the time-space cross-feedback mechanism is:

[0027]

[0028] in, For the The change in the parameter θ of the final intelligent fault detection model in the iteration; and They are The mean and variance of the gradient in the iteration; ∈ is a constant to avoid division by zero; W spatial is the spatial weighting factor; γ3 is the spatiotemporal feedback factor; Num is the total number of input multimodal historical signal feature data; It is The first multimodal historical signal feature data data values; ||·||2 is the L2 norm.

[0029] The beneficial effects of the technical solution of the present invention are:

[0030] 1. By introducing a multi-dimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm, efficient feature extraction of integrated circuit multimodal signal data and multimodal historical signal data is achieved; through a decomposition basis that combines time domain, frequency domain and nonlinear characteristics, the multimodal signal data is decomposed into multiple sub-component signals, each of which can reflect the time-frequency characteristics of the signal; the nonlinear characteristics of the signal are further refined through local nonlinear characteristic analysis, so that the high-order nonlinear dynamics in the signal can be captured; at the same time, a multi-scale dynamic characteristic analysis method is used to perform multi-level processing on each sub-component signal, so that the characteristics of the sub-component signal at different time scales are effectively retained; through this multi-scale dynamic characteristic fusion, a signal representation with high resolution and rich dynamic characteristics is finally obtained.

[0031] 2. Based on multimodal historical signal feature data, an intelligent fault detection model is constructed. The model can realize fault detection of integrated circuits. During the training process, the existing cross-entropy loss function and gradient descent method are optimized to significantly improve the accuracy and robustness of the model. The introduced intelligent fault detection optimization algorithm combines spatiotemporal weighting, nonlinear gradient correction and other technologies to further improve the accuracy and stability of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of an integrated circuit fault detection method based on artificial intelligence described in the present invention. DETAILED DESCRIPTION

[0033] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0035] The specific scheme of an integrated circuit fault detection method based on artificial intelligence provided by the present invention is described in detail below with reference to the accompanying drawings.

[0036] See attached Figure 1 , which shows a flow chart of an integrated circuit fault detection method based on artificial intelligence provided by an embodiment of the present invention, the method comprising the following steps:

[0037] S1. Acquire and preprocess the multimodal signal data and multimodal historical signal data of the integrated circuit to obtain the preprocessed multimodal signal data and multimodal historical signal data, and use the multidimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm to extract features from the preprocessed multimodal signal data and multimodal historical signal data to obtain multimodal signal feature data and multimodal historical signal feature data; construct and train an intelligent fault detection model based on the multimodal historical signal feature data to obtain a final intelligent fault detection model;

[0038] Multimodal signal data in an integrated circuit is acquired using a multimodal sensor, such as a current sensor, a voltage sensor, a temperature sensor, a vibration sensor, etc.; meanwhile, the stored multimodal historical signal data is retrieved from an existing database.

[0039] The multimodal signal data and the multimodal historical signal data are preprocessed such as denoising, normalization, and missing data repair to obtain preprocessed multimodal signal data and preprocessed multimodal historical signal data. The preprocessing process adopts existing technical means and will not be elaborated here.

[0040] The multi-dimensional time-domain-frequency-nonlinear feature fusion extraction algorithm is used to extract features from the pre-processed multi-modal signal data to obtain multi-modal signal feature data. The multi-dimensional time-domain-frequency-nonlinear feature fusion extraction algorithm extracts complex features from the pre-processed multi-modal signal data and introduces adaptive multi-component decomposition, nonlinear dynamic analysis and multi-scale dynamic characteristic fusion in the logical structure to capture the fault mode of multi-modal signals in complex dynamic systems. The specific implementation process is as follows:

[0041] The input signal is the preprocessed multimodal signal data x(t), and the initial analysis time window length is defined as T. In order to perform fine-grained feature extraction on the preprocessed multimodal signal data, a set of decomposition bases that can adapt to multimodal dynamic changes is first constructed. The decomposition base is defined as a composite function that combines the time domain, frequency domain and nonlinear modulation characteristics. The specific expression is:

[0042]

[0043] Among them, Φ n，k (t) is the decomposition basis of the nth layer and the kth frequency band of the preprocessed multimodal signal data at time t. This decomposition basis combines the time domain, frequency domain and nonlinear modulation characteristics; A n is the amplitude coefficient of the decomposition basis, which is used to adjust the intensity of the decomposition basis or the energy distribution of the signal; f n，k is the center frequency of the decomposition basis at the nth layer and the kth frequency band, which represents the frequency characteristics of the preprocessed multimodal signal in the frequency domain; φ kis the phase parameter of the decomposition basis, which is used to adjust the initial phase of the decomposition basis; α n is the time domain resolution factor of the decomposition basis, which is used to control the concentration of the decomposition basis in the time domain; β n is the nonlinear modulation factor of the decomposition basis, which is used to introduce higher-order nonlinear characteristics;

[0044] In the adaptive multi-component decomposition process, based on the role of the decomposition basis, the preprocessed multi-modal signal data is decomposed into multiple sub-component signals x k (t), through integral calculation, the preprocessed multimodal signal data is decomposed into sub-bands in a specific frequency range. The formula for obtaining each sub-component signal is as follows:

[0045] x k (t) = ∫0 T x(t)Φ n，k (t)dt

[0046] Among them, x k (t) is the multimodal signal data x(t) after preprocessing at time t in the frequency subband [f n，k , f n，k+1 ] is the projection within the subcomponent signal; [0, T] is the upper and lower limits of the integration, corresponding to the time window of the initial analysis;

[0047] After the decomposition is completed, a set of subcomponent signals {x1(t), x2(t), ...., x k (t), ..., x K (t)}, these sub-component signals cover the time-frequency characteristics of the preprocessed multimodal signal, and K is the total number of sub-component signals, that is, the total number of frequency bands.

[0048] In order to further capture the nonlinear dynamic characteristics of the signal, for each subcomponent signal x k (t) Perform local nonlinear characteristic analysis. First, calculate the local deviation degree LDD k , the formula is as follows:

[0049]

[0050] Among them, LDD k is the local deviation of the kth subcomponent signal, which is used to measure the deviation of the subcomponent signal from its local mean; is the subcomponent signal x k The local mean of (t) is used to measure the central tendency of the subcomponent signal. The physical meaning of the local deviation degree is to measure whether the deviation degree of the subcomponent signal from its local mean is symmetrical, and the high-order deviation can characterize the local nonlinear characteristics of the signal.

[0051] Further, the nonlinear energy ratio NER is calculatedk , the formula is as follows:

[0052]

[0053] Among them, NER k is the nonlinear energy ratio of the kth subcomponent signal, reflecting the energy proportion of the nonlinear component of the subcomponent signal relative to the preprocessed multimodal signal data; p is the nonlinear order, which is used to improve the sensitivity to high-energy nonlinear components. The nonlinear energy ratio reflects the energy proportion of the subcomponent signal, highlighting the contribution of the nonlinear part of the subcomponent signal to the preprocessed multimodal signal data.

[0054] After obtaining the local nonlinear characteristics of each subcomponent signal, it is mapped into a nonlinear dynamic characteristic matrix M k , used to comprehensively represent the sub-component signal x k (t) nonlinear dynamic characteristics, the matrix contains LDD k and NER k Two indicators:

[0055]

[0056] In order to capture the dynamic characteristics of sub-component signals at different time scales, we further k (t) Perform multi-scale dynamic feature fusion processing and use Gaussian convolution kernel to extract multi-scale dynamic features:

[0057]

[0058] Among them, S k，m (t) is the subcomponent signal x k (t) is the convolution result at the mth scale, which reflects the dynamic characteristics of the sub-component signal at different time scales; σ m is the scale parameter of the Gaussian convolution kernel, which is used to adjust the blurring degree of the sub-component signal at different time scales; t′ represents the integral variable; x k (t′) is the subcomponent signal at time t′;

[0059] For each scale, calculate the energy of the convolution result and use it as the weight:

[0060]

[0061] Among them, wm is the weight of the mth scale, which is used to quantify the proportion of each scale in the total signal energy; M is the total number of scales.

[0062] The final multi-scale dynamic feature fusion is calculated by the following formula:

[0063]

[0064] Among them, F k Represents the subcomponent signal x k (t) Comprehensive dynamic characteristics under multi-scale analysis.

[0065] In order to further characterize the nonlinear interaction between sub-component signals, the nonlinear interaction characteristic matrix is ​​defined Any element is represented by

[0066]

[0067] in, Represents the subcomponent signal x i (t) and x j The nonlinear interaction characteristics between (t) are used to measure the degree of interaction and coupling between the two sub-component signals.

[0068] Finally, the nonlinear dynamic characteristic matrix M of all sub-component signals is combined k , comprehensive dynamic characteristics F k and nonlinear interaction characteristics Generate the comprehensive feature matrix F of the preprocessed multimodal signal data output , F output It is the final feature representation, that is, the multimodal signal feature data, which can comprehensively reflect the multidimensional dynamic characteristics, nonlinear characteristics and nonlinear interaction characteristics of the preprocessed multimodal signal.

[0069] Furthermore, the pre-processed multimodal historical signal data is subjected to feature extraction by the multi-dimensional time-domain and frequency-domain nonlinear feature fusion extraction algorithm to obtain multimodal historical signal feature data F history .

[0070] An intelligent fault detection model is constructed based on multimodal historical signal feature data; the specific process is as follows: a neural network is used as the basic architecture, and at the input layer, multimodal historical signal feature data is received, and weighted processing is performed on the multimodal historical signal feature data to obtain the output of the input layer; and the output of the input layer is used as the input of the hidden layer, and in the hidden layer, a multi-layer perceptron structure is introduced to obtain a multi-layer hidden layer composition. The specific number of layers is determined according to the specific scenario and will not be elaborated here. After processing by the multi-layer hidden layers, the output of the multi-layer hidden layers is used as the input of the output layer to generate the final fault detection result, including faults and non-faults.

[0071] In order to train the intelligent fault detection model, the existing cross entropy loss function is introduced to train the intelligent fault detection model, and the parameters of the intelligent fault detection model are optimized through the existing gradient descent method to obtain the final intelligent fault detection model.

[0072] S2. Optimize the intelligent fault detection model using an intelligent fault detection optimization algorithm to obtain an optimized intelligent fault detection model; process the multimodal signal feature data using the optimized intelligent fault detection model to obtain the fault detection result of the integrated circuit.

[0073] The intelligent fault detection optimization algorithm is used to optimize the intelligent fault detection model to obtain the optimized intelligent fault detection model. The intelligent fault detection optimization algorithm combines technologies such as time-space weighting, nonlinear gradient correction, and multi-level time pane feedback mechanism, which can significantly improve the accuracy and robustness of the optimized intelligent fault detection model. The specific contents include:

[0074] In the training and parameter optimization of the intelligent fault detection model, gradient updates and learning rate adjustments are performed. The traditional gradient descent method is usually used to optimize the loss function L(θ), where θ represents the various parameters in the intelligent fault detection model and is used to control the behavior of the intelligent fault detection model. Optimizing the parameters of the intelligent fault detection model is the core of the training process, and minimizing the loss function is the goal. The update formula of gradient descent is:

[0075]

[0076] in, It is Parameters of the intelligent fault detection model in the iteration; It is The parameters of the intelligent fault detection model in the iteration; η is the learning rate; is the parameter of the current intelligent fault detection model The gradient of the loss function L(θ), that is, the direction and magnitude of the parameter update of the intelligent fault detection model.

[0077] However, due to the complexity of multimodal signals, a single learning rate is often difficult to adapt to the changing speed of different features. In order to better adapt to the diversity of preprocessed multimodal signal data, a dynamic learning rate adjustment mechanism is introduced. In each iteration, the learning rate It gradually decays over time and is dynamically adjusted based on the current gradient fluctuations. The learning rate adjustment formula is:

[0078]

[0079] in, It is The learning rate in the iteration is used to control the step size of each gradient update; η0 is the initial learning rate; λ1, λ2 are parameters that control the decay speed of the learning rate; q, It is the time scale for controlling the attenuation of learning rate; γ1 is the dynamic adjustment factor of learning rate; Used to control the decay of learning rate over time; It is a dynamic adjustment factor used to control the adaptive change of the learning rate. The dynamic learning rate adjustment mechanism can avoid the problem of gradient oscillation caused by too fast learning rate, and also avoid the problem of slow convergence caused by too slow learning rate.

[0080] Nonlinear gradient adjustment and nonlinear activation function are introduced; further, in order to enhance the fitting ability of the intelligent fault detection model to the preprocessed multimodal signal data, a nonlinear activation function is introduced in the gradient update process. The traditional gradient descent method often relies on the linear information of the gradient when updating parameters, which is insufficient for processing complex nonlinear features in multimodal signals. To this end, the gradient update process is corrected by combining nonlinear gradient adjustment with the tanh hyperbolic tangent activation function. The new gradient update formula is:

[0081]

[0082] in, For the The change in the parameter θ of the final intelligent fault detection model in the iteration; and They are The mean and variance of the gradient in the iteration are calculated by an adaptive learning rate method (such as the Adam algorithm); W spatial is a spatial weighting factor, which is used to dynamically adjust the weight of each feature and to indicate the contribution of different features to the performance of the intelligent fault detection model; ∈ is a constant to avoid division by zero; γ2 is an adjustment factor, which determines the nonlinear correction strength of the nonlinear activation function, i.e., the tanh hyperbolic tangent activation function, on the gradient; Num is the total number of input multimodal historical signal feature data; It is The multimodal historical signal feature data of the input is The mean of the gradients in iterations; It is The multimodal historical signal feature data of the input is The variance of the gradient in iteration .

[0083] The purpose of introducing the tanh hyperbolic tangent activation function is to correct the gradient through nonlinear transformation, thereby avoiding the gradient explosion or gradient vanishing problem, and at the same time enhancing the ability of the intelligent fault detection model to handle complex nonlinear relationships.

[0084] Since different multimodal historical signal feature data have different contributions in fault detection, feature selection and weighting are important steps in the optimization process. The spatial weighting mechanism is introduced to weight and update each multimodal historical signal feature data through the spatial weighting factor to improve the performance of the intelligent fault detection model. The calculation formula of the spatial weighting factor is:

[0085]

[0086] in, It is The first multimodal historical signal feature data data values; ||·||2 is the L2 norm, which indicates the size of the multimodal historical signal feature data; is a weighting coefficient used to control the adjustment strength of feature importance; ∈ is a constant to avoid division by zero; Num is the total number of input multimodal historical signal feature data; is the total number of data values ​​of the multimodal historical signal feature data. The purpose of the spatial weighting mechanism is to dynamically adjust the weight of each multimodal historical signal feature data according to its importance in different spatial dimensions, thereby improving the sensitivity of the intelligent fault detection model to the multimodal historical signal feature data.

[0087] Introduce the spatiotemporal cross-feedback mechanism; Considering the spatiotemporal characteristics of multimodal historical signal feature data, the features not only depend on the time dimension, but are also affected by the space dimension, so the spatiotemporal cross-feedback mechanism is introduced. The spatiotemporal cross-feedback mechanism combines the information in the time and space dimensions and further optimizes the learning process of the intelligent fault detection model. The expression of the spatiotemporal cross-feedback mechanism is:

[0088]

[0089] in, For the The change in the parameter θ of the final intelligent fault detection model in the iteration; γ3 is the spatiotemporal feedback factor, which is used to control the feedback effect between temporal and spatial characteristics.

[0090] Through the spatiotemporal cross-feedback mechanism, the intelligent fault detection model can comprehensively consider the relationship between temporal and spatial features in each iteration, thereby improving the accuracy of fault detection.

[0091] After the above processing, the update formula of the optimized gradient descent is:

[0092]

[0093] in, It is Parameters of the intelligent fault detection model in the iteration; It is Parameters of the intelligent fault detection model in the iteration;

[0094] After the above optimization processing, an optimized intelligent fault detection model is obtained.

[0095] Furthermore, the optimized intelligent fault detection model is used to process the multimodal signal feature data to obtain the fault detection results of the integrated circuit, including faults and non-faults.

[0096] In summary, an integrated circuit fault detection method based on artificial intelligence is completed.

[0097] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0098] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0099] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. An integrated circuit fault detection method based on artificial intelligence, characterized in that: The following steps are involved: S1. Acquire and preprocess the multimodal signal data and multimodal historical signal data of the integrated circuit to obtain the preprocessed multimodal signal data and multimodal historical signal data, and use the multidimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm to extract features from the preprocessed multimodal signal data and multimodal historical signal data to obtain multimodal signal feature data and multimodal historical signal feature data; Based on the multimodal historical signal feature data, an intelligent fault detection model is constructed and trained to obtain the final intelligent fault detection model; S2. Optimize the intelligent fault detection model using an intelligent fault detection optimization algorithm to obtain an optimized intelligent fault detection model; process the multimodal signal feature data using the optimized intelligent fault detection model to obtain the fault detection result of the integrated circuit.

2. The integrated circuit fault detection method based on artificial intelligence according to claim 1, characterized in that: The S1 specifically includes: The multi-dimensional time-domain-frequency-nonlinear feature fusion extraction algorithm extracts features from pre-processed multi-modal signal data and multi-modal historical signal data by introducing adaptive multi-component decomposition, nonlinear dynamics analysis and multi-scale dynamic characteristic fusion.

3. The integrated circuit fault detection method based on artificial intelligence according to claim 2 is characterized in that: The S1 specifically includes: In the implementation process of the multidimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm, the decomposition basis is constructed by combining the time domain, frequency domain and nonlinear modulation characteristics; in the adaptive multi-component decomposition process, based on the role of the decomposition basis, the preprocessed multimodal signal data is decomposed to obtain sub-component signals.

4. The integrated circuit fault detection method based on artificial intelligence according to claim 3 is characterized in that: The S1 specifically includes: In the implementation process of the multidimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm, the local nonlinear characteristics of each sub-component signal are analyzed to obtain the local deviation degree and nonlinear energy ratio of the sub-component signal, and the local deviation degree and nonlinear energy ratio of the sub-component signal are mapped into a nonlinear dynamic characteristic matrix.

5. The integrated circuit fault detection method based on artificial intelligence according to claim 4 is characterized in that: The S1 specifically includes: In the implementation process of the multi-dimensional time-domain and frequency-domain nonlinear feature fusion extraction algorithm, multi-scale dynamic characteristic fusion processing is performed on each sub-component signal, and a Gaussian convolution kernel is introduced to extract the multi-scale dynamic characteristics. The convolution results of the sub-component signal at different scales are obtained, and the energy of the convolution result is calculated as the weight of each scale, and finally the comprehensive dynamic characteristics of the sub-component signal are obtained.

6. The integrated circuit fault detection method based on artificial intelligence according to claim 5, characterized in that: The S1 specifically includes: In the implementation process of the multidimensional time domain-frequency domain-nonlinear feature fusion extraction algorithm, a nonlinear interaction characteristic matrix is ​​defined to determine the nonlinear interaction between sub-component signals; and the nonlinear dynamic characteristic matrix, comprehensive dynamic characteristics and nonlinear interaction characteristics of the sub-component signals are combined to obtain a comprehensive characteristic matrix, that is, multimodal signal feature data.

7. The integrated circuit fault detection method based on artificial intelligence according to claim 1, characterized in that: The S2 specifically includes: In the intelligent fault detection optimization algorithm, the intelligent fault detection model is updated with gradients and the learning rate is adjusted. By introducing a dynamic learning rate adjustment mechanism, the learning rate is dynamically adjusted in each iteration in combination with the fluctuation of the current gradient. The specific implementation formula is: in, It is The learning rate in the iteration; η0 is the initial learning rate; λ1, λ2 are parameters that control the decay speed of the learning rate; q, is the time scale for controlling the attenuation of the learning rate; γ1 is the factor for dynamically adjusting the learning rate.

8. The integrated circuit fault detection method based on artificial intelligence according to claim 7, characterized in that: The S2 specifically includes: In the intelligent fault detection optimization algorithm, the gradient update process is corrected by combining nonlinear gradient adjustment with nonlinear activation function. The spatial weighting mechanism is introduced to perform weighted update on each multimodal historical signal feature data through the spatial weighting factor.

9. The integrated circuit fault detection method based on artificial intelligence according to claim 8, characterized in that: The S2 specifically includes: In the intelligent fault detection optimization algorithm, the time-space cross-feedback mechanism is introduced to optimize the learning process of the intelligent fault detection model, and finally the optimized intelligent fault detection model is obtained. The expression of the time-space cross-feedback mechanism is: in, For the The change in the parameter θ of the final intelligent fault detection model in the iteration; and They are The mean and variance of the gradient in the iteration; ∈ is a constant to avoid division by zero; W spatial is the spatial weighting factor; γ3 is the spatiotemporal feedback factor; Num is the total number of input multimodal historical signal feature data; It is The first multimodal historical signal feature data data values; ||·||2 is the L2 norm.

Images