Artificial intelligence-based pharmaceutical equipment fault diagnosis prediction method and system

Abstract

The application relates to the technical field of pharmaceutical equipment, and discloses a pharmaceutical equipment fault diagnosis and prediction method and system based on artificial intelligence, which comprises the following steps: signal conditioning is performed on multi-source operation data collected by pharmaceutical equipment to obtain standardized time series data of the pharmaceutical equipment; adaptive blind source separation is performed on the standardized time series data to obtain entity state features and process execution features of the pharmaceutical equipment; fault mechanism tracing is performed on the pharmaceutical equipment to obtain a fault propagation path of the pharmaceutical equipment; correlation evaluation is performed on the process execution features to obtain health state representation of the pharmaceutical equipment; the health state representation is mapped to historical equipment service data of the pharmaceutical equipment to predict the remaining service life of the pharmaceutical equipment, and a risk time window of the pharmaceutical equipment is obtained; and the maintenance task and the production plan of the pharmaceutical equipment are cooperatively optimized to obtain a predictive maintenance work order of the pharmaceutical equipment; and the application can improve the efficiency of the fault diagnosis and prediction of the pharmaceutical equipment based on artificial intelligence.

Description

Technical Field

[0001] This invention relates to the field of pharmaceutical equipment technology, and in particular to a method and system for fault diagnosis and prediction of pharmaceutical equipment based on artificial intelligence. Background Technology

[0002] Pharmaceutical equipment generates multi-source heterogeneous operational data during operation. Existing technologies lack standardized signal conditioning processes for processing this type of data, fail to achieve synchronization and dimensional optimization of multi-source data, and struggle to effectively extract effective features from the data. Furthermore, the separation methods for features lack adaptability, making it impossible to accurately distinguish between the physical state characteristics of pharmaceutical equipment and process execution characteristics. This results in insufficient basic data support for subsequent fault diagnosis and poor accuracy and relevance of feature analysis.

[0003] Current technologies for fault diagnosis in pharmaceutical equipment mostly only identify single faults, failing to trace fault mechanisms, explore fault propagation paths, or correlate fault characteristics with process execution characteristics. This results in an inability to accurately characterize equipment health status. Furthermore, the prediction of remaining equipment lifespan lacks precise mapping to historical service data, making it difficult to determine reasonable risk time windows. Moreover, the technologies fail to achieve coordinated optimization between maintenance tasks and production plans, resulting in insufficient comprehensiveness and practicality in fault prediction, and poor scientific rigor and rationality in maintenance decisions. Therefore, improving the accuracy of fault diagnosis, the foresight of fault prediction, and the coordination between maintenance and production plans have become urgent problems to be solved. Summary of the Invention

[0004] This invention provides a method and system for fault diagnosis and prediction of pharmaceutical equipment based on artificial intelligence, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides an artificial intelligence-based method for fault diagnosis and prediction of pharmaceutical equipment, comprising: S1. Perform signal conditioning on the multi-source operating data collected from the pharmaceutical equipment to obtain standardized time-series data of the pharmaceutical equipment. The multi-source operating data includes vibration signals, temperature data, pressure data, and motor current data. S2. Adaptive blind source separation is performed on standardized time-series data to obtain the physical state characteristics and process execution characteristics of pharmaceutical equipment; S3. Based on the ontological state characteristics, trace the fault mechanism of pharmaceutical equipment to obtain the fault propagation path of pharmaceutical equipment. S4. Based on the fault propagation path, the process execution characteristics are correlated and evaluated to obtain the health status characterization of the pharmaceutical equipment. S5. Map the health status characteristics to the historical equipment service data of pharmaceutical equipment to predict the remaining service life of pharmaceutical equipment and obtain the risk time window of pharmaceutical equipment. S6. Based on the risk time window, the maintenance tasks and production plans of pharmaceutical equipment are optimized in a coordinated manner to obtain predictive maintenance work orders for pharmaceutical equipment.

[0006] In a preferred embodiment, the multi-source operational data collected from the pharmaceutical equipment is signal conditioned to obtain standardized time-series data of the pharmaceutical equipment. The multi-source operational data includes vibration signals, temperature data, pressure data, and motor current data, including: By collecting vibration signals, temperature data, pressure data, and motor current data of pharmaceutical equipment, multi-source operating data of the pharmaceutical equipment can be obtained. Sliding window autocorrelation analysis was performed on multi-source operational data to obtain the intra-source periodicity characteristics of pharmaceutical equipment; Cross-correlation mapping of periodic features within the source yields the time delay relationship of multi-source operational data; Based on the time delay relationship, phase compensation is performed on temperature data, pressure data, and motor current data to obtain synchronous multi-source data of pharmaceutical equipment. By reducing the intrinsic dimension of the synchronous multi-source data, the principal component feature matrix of pharmaceutical equipment is obtained. The principal component feature matrix is ​​normalized and registered to obtain standardized time-series data of pharmaceutical equipment.

[0007] In a preferred embodiment, the adaptive blind source separation of standardized time-series data to obtain the physical state characteristics and process execution characteristics of the pharmaceutical equipment includes: The standardized time-series data were subjected to decorrelation processing to obtain the whitening data matrix of pharmaceutical equipment; The covariance matrix of the whitened data matrix is ​​orthogonally decomposed, and the decomposed eigenvalues ​​are arranged in descending order to obtain the eigenvalue spectrum sequence of pharmaceutical equipment. The eigenvalue spectrum sequence is ordered by successive drop ratios to obtain the source signal dimension of the pharmaceutical equipment. Based on the source signal dimension, source separation is performed on the whitened data matrix to obtain the independent source signal components of the pharmaceutical equipment; Power spectral density analysis is performed on independent source signal components to distinguish between high-frequency impact components and low-frequency trend components, and these components are used as the physical state characteristics and process execution characteristics of pharmaceutical equipment.

[0008] In a preferred embodiment, the step of determining the order of the eigenvalue spectrum sequence by successive drop ratios to obtain the source signal dimension of the pharmaceutical equipment includes: The eigenvalues ​​in the eigenvalue spectrum sequence are paired up step by step to obtain the eigenvalue pairs of the eigenvalue spectrum sequence. The drop ratio of each successive eigenvalue pair is calculated to obtain the single-order drop ratio of the successive eigenvalue pairs; the formula for calculating the drop ratio is as follows: ; in, Indicates the first The single-level drop ratio of the steps, Indicates the first 1 eigenvalue, Indicates the order of descending order. 1 eigenvalue, Indicates the order of the eigenvalues. The range of values ​​is to , This represents the total number of eigenvalues ​​in the eigenvalue spectrum sequence; The single-order drop ratio is serialized and recombined to obtain the drop ratio sequence of the eigenvalue spectrum sequence; The order corresponding to the maximum value in the drop ratio sequence is taken as the dimension of the source signal of the pharmaceutical equipment.

[0009] In a preferred embodiment, the step of tracing the fault mechanism of pharmaceutical equipment based on its physical state characteristics to obtain the fault propagation path of the pharmaceutical equipment includes: Empirical wavelet decomposition is performed on the ontological state features to obtain the modal components of the pharmaceutical equipment. The information transmission direction and intensity between quantified modal components are used to construct an information flow matrix for pharmaceutical equipment. A fault information transmission network for pharmaceutical equipment is constructed using modal components as nodes and the transmission entropy value in the information flow matrix as the weight of the directed edges. Using the reciprocal of the transmission entropy as the path weight, the shortest path tracing is performed on the fault information transmission network to obtain the fault propagation path of the pharmaceutical equipment.

[0010] In a preferred embodiment, the step of performing correlation assessment on process execution characteristics based on fault propagation paths to obtain a health status characterization of pharmaceutical equipment includes: The process execution characteristics are extracted to obtain the process fluctuation trajectory of the pharmaceutical equipment; Based on the fault propagation path, node mapping and segmentation are performed on the process fluctuation trajectory to obtain the path fluctuation correlation segment of pharmaceutical equipment. By performing trend consistency judgment on the path fluctuation correlation segments, the deterioration transmission trend of pharmaceutical equipment along the fault propagation path can be obtained. By calibrating the health level of the degradation transmission trend, the health status of the pharmaceutical equipment can be characterized.

[0011] In a preferred embodiment, the step of mapping health status characteristics to historical equipment service data of pharmaceutical equipment to predict the remaining service life of pharmaceutical equipment and obtain the risk time window of pharmaceutical equipment includes: By jointly annotating the historical service data of pharmaceutical equipment, a sample library of historical healthy lifespan of pharmaceutical equipment is obtained. Centroid clustering was performed on the historical health status of the historical health life sample database to obtain a health status benchmark template for pharmaceutical equipment. Based on the health status benchmark template, the nearest neighbor search is performed on the health status representation to obtain the matching health status template of the pharmaceutical equipment. Based on the matching health status template, historical remaining life samples associated with the matching health status template are indexed from the historical health life sample library, and interval reduction is performed on the historical remaining life samples to obtain the remaining life prediction interval of pharmaceutical equipment. The remaining life prediction interval is defined as the risk time window for pharmaceutical equipment.

[0012] In a preferred embodiment, the step of performing nearest neighbor retrieval on the health status representation based on the health status benchmark template to obtain a matching health status template for the pharmaceutical equipment includes: A similarity space mapping is performed between the health status representation and the health status benchmark template to obtain the template difference sequence between the health status representation and the health status benchmark template; The minimum difference value in the template difference sequence is extracted as the nearest response value of the health status baseline template, and the nearest response value is marked with a position index to obtain the template index corresponding to the nearest response value. Use the health status benchmark template pointed to by the template index as the matching health status template for pharmaceutical equipment.

[0013] In a preferred embodiment, the step of co-optimizing the maintenance tasks and production plans of pharmaceutical equipment based on a risk time window to obtain predictive maintenance work orders for the pharmaceutical equipment includes: By detecting the overlap between the risk time window and the production plan of pharmaceutical equipment, a set of conflicting production tasks for pharmaceutical equipment is obtained. Based on the conflict production task set, the urgency of the maintenance tasks to be performed on the pharmaceutical equipment is assigned to obtain a priority list of maintenance tasks for the pharmaceutical equipment. Based on the maintenance task priority list, the production periods in the production plan are rearranged in sequence to obtain an optimized production plan for pharmaceutical equipment. By integrating and arranging maintenance tasks from the optimized production plan and maintenance task priority list, predictive maintenance work orders for pharmaceutical equipment are obtained.

[0014] To address the above problems, the present invention also provides an artificial intelligence-based fault diagnosis and prediction system for pharmaceutical equipment, the system comprising: The method of co-optimizing maintenance tasks and production plans for pharmaceutical equipment based on risk time windows to obtain predictive maintenance work orders for pharmaceutical equipment includes: By detecting the overlap between the risk time window and the production plan of pharmaceutical equipment, a set of conflicting production tasks for pharmaceutical equipment is obtained. Based on the conflict production task set, the urgency of the maintenance tasks to be performed on the pharmaceutical equipment is assigned to obtain a priority list of maintenance tasks for the pharmaceutical equipment. Based on the maintenance task priority list, the production periods in the production plan are rearranged in sequence to obtain an optimized production plan for pharmaceutical equipment. By integrating and arranging maintenance tasks from the optimized production plan and maintenance task priority list, predictive maintenance work orders for pharmaceutical equipment are obtained.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This technology performs standardized signal conditioning and adaptive blind source separation on multi-source operating data of pharmaceutical equipment to accurately extract the equipment's physical state characteristics and process execution characteristics. Then, based on the physical state characteristics, it completes fault mechanism tracing and explores fault propagation paths. At the same time, it combines the fault propagation paths with the process execution characteristics for correlation evaluation, thereby achieving accurate characterization of equipment health status. This significantly improves the accuracy of fault diagnosis and the pertinence of feature analysis in pharmaceutical equipment, making the identification and tracing of equipment faults more scientific and enabling the grasp of abnormal operating conditions of equipment from the root.

[0016] 2. This technology accurately maps equipment health status characteristics with historical service data, enabling precise prediction of remaining service life and delineation of risk time windows. Based on these risk time windows, it also optimizes maintenance tasks and production plans collaboratively and generates predictive maintenance work orders. This effectively improves the foresight of pharmaceutical equipment failure prediction and the rationality of maintenance decisions, achieving efficient collaboration between equipment maintenance and production operations. It not only ensures the stable operation of pharmaceutical equipment but also maximizes the utilization of production resources, improving the management efficiency of the entire equipment lifecycle and the continuity of pharmaceutical production. Attached Figure Description

[0017] Figure 1 A flowchart illustrating an artificial intelligence-based pharmaceutical equipment fault diagnosis and prediction method provided in an embodiment of the present invention; Figure 2 A functional block diagram of an artificial intelligence-based pharmaceutical equipment fault diagnosis and prediction system provided in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides an artificial intelligence-based method for fault diagnosis and prediction of pharmaceutical equipment. The executing entity of this AI-based method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the AI-based method for fault diagnosis and prediction of pharmaceutical equipment can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating an artificial intelligence-based fault diagnosis and prediction method for pharmaceutical equipment according to an embodiment of the present invention. In this embodiment, the artificial intelligence-based fault diagnosis and prediction method for pharmaceutical equipment includes: S1. Perform signal conditioning on the multi-source operating data collected from the pharmaceutical equipment to obtain standardized time-series data of the pharmaceutical equipment. The multi-source operating data includes vibration signals, temperature data, pressure data, and motor current data. In this embodiment of the invention, the signal conditioning of the multi-source operating data collected from the pharmaceutical equipment to obtain standardized time-series data of the pharmaceutical equipment, the multi-source operating data including vibration signals, temperature data, pressure data, and motor current data, includes: By collecting vibration signals, temperature data, pressure data, and motor current data of pharmaceutical equipment, multi-source operating data of the pharmaceutical equipment can be obtained. Sliding window autocorrelation analysis was performed on multi-source operational data to obtain the intra-source periodicity characteristics of pharmaceutical equipment; Cross-correlation mapping of periodic features within the source yields the time delay relationship of multi-source operational data; Based on the time delay relationship, phase compensation is performed on temperature data, pressure data, and motor current data to obtain synchronous multi-source data of pharmaceutical equipment. By reducing the intrinsic dimension of the synchronous multi-source data, the principal component feature matrix of pharmaceutical equipment is obtained. The principal component feature matrix is ​​normalized and registered to obtain standardized time-series data of pharmaceutical equipment.

[0021] Vibration signals, temperature data, pressure data, and motor current data are extracted from various monitoring sensors of the pharmaceutical equipment. These data are then integrated and summarized according to the time dimension of collection to form multi-source operating data of the pharmaceutical equipment that includes all monitoring types. During the integration process, the original collection timestamps and data values ​​of each type of data are retained to ensure data integrity.

[0022] Multi-source operational data of pharmaceutical equipment is segmented according to a set time window length. Correlation analysis is performed on the single type of data within each time window. By analyzing the degree of numerical correlation of data at different time points, the periodic change pattern of each type of data is extracted, thereby obtaining the intra-source periodic characteristics of pharmaceutical equipment. This characteristic can accurately reflect the repetitive pattern of single data type changes over time.

[0023] The extracted periodic features of various data types are used to perform pairwise correlation mapping analysis. The correlation between the periodic features of different data types on the time axis is analyzed to clarify the time delay between the periodic changes of one type of data and the periodic changes of another type of data. Through comprehensive correlation analysis, the overall time delay relationship of multi-source running data is obtained, and the matching differences of various types of data in the time dimension are clearly presented.

[0024] Based on the time delay relationship of the obtained multi-source operating data, the timestamps of temperature data, pressure data, and motor current data are adjusted and corrected. Data with time delays are calibrated forward or backward according to the time delay difference, so that the temperature data, pressure data, motor current data and vibration signal are completely synchronized in the time dimension, and the values ​​of various data at the same point in time can be matched accordingly, ultimately obtaining the synchronous multi-source data of pharmaceutical equipment.

[0025] Feature extraction and dimensional simplification are performed on the synchronized multi-source data of pharmaceutical equipment that has completed time synchronization. Redundant information and irrelevant features in the data are removed, and core feature information that can reflect the operating status of pharmaceutical equipment is retained. The core features are arranged in a matrix according to data type and time dimension to form a clear row and column correspondence of pharmaceutical equipment principal component feature matrix. The values ​​in the matrix only retain the key operating features of the multi-source data.

[0026] The principal component feature matrix of pharmaceutical equipment is numerically standardized and adjusted to map feature values ​​with different dimensions and numerical ranges within the matrix to the same numerical range. During the adjustment process, the row and column structure of the feature matrix and the correspondence between features remain unchanged, so that all values ​​within the matrix are comparable and analyzable, and finally standardized time series data of pharmaceutical equipment are obtained.

[0027] The beneficial effects are that through this series of signal conditioning steps, the effective collection and time synchronization of multi-source operating data of pharmaceutical equipment are realized, solving the time delay problem caused by the acquisition characteristics of multi-source data. At the same time, the data dimensionality is simplified and the values ​​are standardized, redundant information is eliminated, core operating characteristics are retained, and the standardized time series data formed can provide unified, standardized and high-quality basic data support for subsequent feature separation and fault diagnosis, effectively improving the accuracy and efficiency of subsequent data processing and analysis.

[0028] S2. Adaptive blind source separation is performed on standardized time-series data to obtain the physical state characteristics and process execution characteristics of pharmaceutical equipment; In this embodiment of the invention, the adaptive blind source separation of standardized time-series data to obtain the physical state characteristics and process execution characteristics of pharmaceutical equipment includes: The standardized time-series data were subjected to decorrelation processing to obtain the whitening data matrix of pharmaceutical equipment; The covariance matrix of the whitened data matrix is ​​orthogonally decomposed, and the decomposed eigenvalues ​​are arranged in descending order to obtain the eigenvalue spectrum sequence of pharmaceutical equipment. The eigenvalue spectrum sequence is ordered by successive drop ratios to obtain the source signal dimension of the pharmaceutical equipment. Based on the source signal dimension, source separation is performed on the whitened data matrix to obtain the independent source signal components of the pharmaceutical equipment; Power spectral density analysis is performed on independent source signal components to distinguish between high-frequency impact components and low-frequency trend components, and these components are used as the physical state characteristics and process execution characteristics of pharmaceutical equipment.

[0029] The step of determining the order of the eigenvalue spectrum sequence by successive drop ratios to obtain the source signal dimension of the pharmaceutical equipment includes: The eigenvalues ​​in the eigenvalue spectrum sequence are paired up step by step to obtain the eigenvalue pairs of the eigenvalue spectrum sequence. The drop ratio of each successive eigenvalue pair is calculated to obtain the single-order drop ratio of the successive eigenvalue pairs; the formula for calculating the drop ratio is as follows: ; in, Indicates the first The single-level drop ratio of the steps, Indicates the first 1 eigenvalue, Indicates the order of descending order. 1 eigenvalue, Indicates the order of the eigenvalues. The range of values ​​is to , This represents the total number of eigenvalues ​​in the eigenvalue spectrum sequence; The single-order drop ratio is serialized and recombined to obtain the drop ratio sequence of the eigenvalue spectrum sequence; The order corresponding to the maximum value in the drop ratio sequence is taken as the dimension of the source signal of the pharmaceutical equipment.

[0030] The standardized time-series data of pharmaceutical equipment is subjected to decorrelation processing to sort out the linear correlation information between the various dimensions of the data. The correlation between different dimensional features is eliminated through data transformation, so that the processed data presents an independent state for each dimension. At the same time, the variance of the data is adjusted to a fixed uniform value so that the data distribution meets the core requirements of whitening processing. Finally, a whitening data matrix of pharmaceutical equipment with regular structure and independent dimensions is formed.

[0031] A covariance matrix is ​​constructed for the whitening data matrix of pharmaceutical equipment. This matrix can accurately reflect the dispersion and correlation characteristics of the data in each dimension of the whitening data matrix. Orthogonal eigenvalue decomposition is performed on the constructed covariance matrix to obtain the corresponding eigenvalues ​​and eigenvectors. All the eigenvalues ​​obtained by decomposition are arranged in descending order of value. During the arrangement process, the original attributes and related information of each eigenvalue are completely preserved, and finally an ordered sequence of eigenvalues ​​of pharmaceutical equipment is formed.

[0032] Adjacent feature values ​​in the feature value spectrum sequence of pharmaceutical equipment are paired sequentially. Starting from the first feature value in the sequence, the previous feature value is paired with the next feature value in turn until the last feature value in the sequence is paired. Each pair retains the original value of the feature value and its position information in the sequence, and finally obtains the successive feature value pairs of the feature value spectrum sequence of pharmaceutical equipment.

[0033] For each pair of successive eigenvalues ​​in the eigenvalue spectrum sequence of pharmaceutical equipment, the drop ratio is calculated. Numerical operations are performed on each pair of eigenvalues, and the results can accurately reflect the degree of numerical drop between the two eigenvalues ​​in each pair, providing data support for subsequent order determination operations. Finally, the single-order drop ratio of the successive eigenvalue pairs in the eigenvalue spectrum sequence of pharmaceutical equipment is obtained.

[0034] No. The eigenvalue and the descending order of the eigenvalue Each eigenvalue comes from the eigenvalue spectrum sequence of the pharmaceutical equipment obtained by orthogonally decomposing the covariance matrix of the whitened data matrix of the pharmaceutical equipment and arranging it in descending order. The order of the eigenvalues ​​is... The order is determined by the arrangement of eigenvalues ​​in the eigenvalue spectrum sequence. The total number of eigenvalues ​​in the eigenvalue spectrum sequence is the total number of eigenvalues ​​contained in the eigenvalue spectrum sequence. The range of values ​​is limited to from The consecutive integers up to the total number of eigenvalues ​​minus one, the th The single-order drop ratio is the result obtained by operating on two eigenvalues ​​of the corresponding order. The core significance of this calculation method lies in quantifying the degree of numerical difference between two adjacent eigenvalues ​​in the eigenvalue spectrum sequence. By visually reflecting the magnitude of numerical change at each adjacent position in the eigenvalue spectrum sequence through the operation results of adjacent eigenvalues, it provides a quantifiable basis for determining the dimension of the source signal of pharmaceutical equipment, accurately reflecting the distribution differences and variation characteristics of eigenvalues ​​of each order in the eigenvalue spectrum sequence. The greater the numerical difference between two adjacent eigenvalues ​​in the eigenvalue spectrum sequence, the larger the single-order drop ratio calculated by this method; conversely, the closer the values ​​of two adjacent eigenvalues, the smaller the single-order drop ratio. When the value of the preceding eigenvalue in the eigenvalue spectrum sequence is slightly greater than that of the following eigenvalue, the single-order drop ratio shows a small positive correlation; when the value of the preceding eigenvalue in the eigenvalue spectrum sequence is much greater than that of the following eigenvalue, the single-order drop ratio shows a significant increase.

[0035] All single-order drop ratios of the eigenvalue spectrum sequence of pharmaceutical equipment are rearranged and reordered according to their corresponding order-wise eigenvalues. During the reordering process, the value of each single-order drop ratio and its corresponding order information are retained to form an ordered numerical sequence, and finally the drop ratio sequence of the eigenvalue spectrum sequence of pharmaceutical equipment is obtained.

[0036] From the drop ratio sequence of the eigenvalue spectrum sequence of pharmaceutical equipment, the single-order drop ratio with the largest value is selected, and the permutation order corresponding to the largest value in the drop ratio sequence is accurately determined. This order is directly defined as the source signal dimension of the pharmaceutical equipment, thus defining a clear dimensional range for subsequent source separation.

[0037] Based on the determined source signal dimension of the pharmaceutical equipment, source separation processing is carried out on the whitened data matrix of the pharmaceutical equipment. According to the dimensional boundary defined by the source signal dimension, signal components that are not related to each other are separated from the whitened data matrix. Each signal component retains its own independent features and numerical information, with no feature overlap, and finally the independent source signal components of the pharmaceutical equipment are obtained.

[0038] Power spectral density analysis was performed on the independent source signal components of pharmaceutical equipment to analyze the power spectral density distribution characteristics and frequency variation patterns of each independent source signal component. The independent source signal components were accurately divided according to the high and low frequency ranges of the signal. Signal components with high frequencies and obvious impact change characteristics were selected as high-frequency impact components, and signal components with low frequencies and stable trend change characteristics were selected as low-frequency trend components. The high-frequency impact components were directly determined as the physical state characteristics of the pharmaceutical equipment, and the low-frequency trend components were directly determined as the process execution characteristics of the pharmaceutical equipment.

[0039] The beneficial effects are as follows: by carrying out adaptive blind source separation processing on standardized time-series data throughout the entire process, a progressive processing of data decorrelation, feature decomposition, order determination analysis, and source separation is achieved. The dimension of the source signal is accurately determined, ensuring the pertinence and accuracy of source separation. Furthermore, power spectral density analysis enables precise differentiation between the body state characteristics and process execution characteristics, making the two characteristics independent and their attributes highly matched with the equipment operating status. This provides accurate, independent, and targeted feature data support for subsequent fault mechanism tracing based on body state characteristics and health status assessment based on process execution characteristics, effectively improving the scientific nature, accuracy, and efficiency of subsequent fault diagnosis and status assessment work.

[0040] S3. Based on the ontological state characteristics, trace the fault mechanism of pharmaceutical equipment to obtain the fault propagation path of pharmaceutical equipment. In this embodiment of the invention, the step of tracing the fault mechanism of pharmaceutical equipment based on its ontological state characteristics to obtain the fault propagation path of the pharmaceutical equipment includes: Empirical wavelet decomposition is performed on the ontological state features to obtain the modal components of the pharmaceutical equipment. The information transmission direction and intensity between quantified modal components are used to construct an information flow matrix for pharmaceutical equipment. A fault information transmission network for pharmaceutical equipment is constructed using modal components as nodes and the transmission entropy value in the information flow matrix as the weight of the directed edges. Using the reciprocal of the transmission entropy as the path weight, the shortest path tracing is performed on the fault information transmission network to obtain the fault propagation path of the pharmaceutical equipment.

[0041] Empirical wavelet decomposition is performed on the physical state characteristics of pharmaceutical equipment. First, reasonable frequency intervals are divided according to the signal characteristics of the physical state characteristics. Then, corresponding empirical wavelet basis functions are constructed for each frequency interval. The signals of the physical state characteristics are projected onto each empirical wavelet basis function to decompose the original physical state characteristics into multiple independent signal components with different frequency scales. Each signal component retains the characteristic information of the corresponding frequency, and finally the modal components of pharmaceutical equipment are obtained.

[0042] A quantitative analysis of the information transmission characteristics of each modal component of pharmaceutical equipment is performed. By analyzing the dynamic changes and correlations between different modal components, the specific direction of information transmission between each pair of modal components is clarified. At the same time, the strength of information transmission in this direction is numerically calibrated. The quantitative results of the information transmission direction and intensity of all modal components are arranged in a matrix, with the rows and columns of the matrix corresponding to the modal components. The numerical values ​​in the matrix represent the transmission entropy values ​​between the modal components, and finally, the information flow matrix of pharmaceutical equipment is constructed.

[0043] Each modal component of the pharmaceutical equipment is treated as an independent node. The transfer entropy value between each modal component in the information flow matrix of the pharmaceutical equipment is extracted. This transfer entropy value is used as the weight of the directed edge connecting the corresponding node. The direction of the directed edge is set according to the information transmission direction represented in the information flow matrix. All nodes are connected and integrated with the weighted directed edges to form a network structure that can fully reflect the information transmission relationship between modal components. Finally, the fault information transmission network of the pharmaceutical equipment is constructed.

[0044] All transmission entropy values ​​in the information flow matrix of pharmaceutical equipment are converted into their inverses. The converted values ​​are used as the path weights of the corresponding directed edges in the fault information transmission network of pharmaceutical equipment. Based on these path weights, path retrieval is performed in the fault information transmission network. Starting from the starting node of the network, all possible paths are traversed, the total weight of each complete path is calculated and compared, and the path with the smallest total weight is selected. This path is the critical path from the generation of the fault to its propagation, and finally, the fault propagation path of the pharmaceutical equipment is obtained.

[0045] The beneficial effects are as follows: through this series of fault mechanism tracing steps, the body state characteristics are refined by using empirical wavelet decomposition. The fault information transmission network constructed by quantifying information transmission characteristics can accurately reflect the fault correlation within the equipment. Then, based on the path weight of the inverse of the transmission entropy value, the shortest path tracing is completed, which can accurately locate the propagation path of pharmaceutical equipment faults. This achieves in-depth mining from the body state characteristics to the fault propagation mechanism, allowing fault diagnosis to extend from simple anomaly identification to the root cause analysis of fault propagation. It provides accurate fault propagation basis for subsequent health status assessment combined with process execution characteristics, and greatly improves the depth and scientificity of fault diagnosis.

[0046] S4. Based on the fault propagation path, the process execution characteristics are correlated and evaluated to obtain the health status characterization of the pharmaceutical equipment. In this embodiment of the invention, the step of performing correlation evaluation on process execution characteristics based on fault propagation paths to obtain a health status characterization of pharmaceutical equipment includes: The process execution characteristics are extracted to obtain the process fluctuation trajectory of the pharmaceutical equipment; Based on the fault propagation path, node mapping and segmentation are performed on the process fluctuation trajectory to obtain the path fluctuation correlation segment of pharmaceutical equipment. By performing trend consistency judgment on the path fluctuation correlation segments, the deterioration transmission trend of pharmaceutical equipment along the fault propagation path can be obtained. By calibrating the health level of the degradation transmission trend, the health status of the pharmaceutical equipment can be characterized.

[0047] The process execution characteristics of pharmaceutical equipment are analyzed in all dimensions. All numerical change points of the process execution characteristics during operation are extracted according to the time dimension. The change points are connected continuously according to the time sequence to fully restore the numerical fluctuation process of the process execution characteristics over time. The numerical fluctuation state of the process execution characteristics at different time stages is clearly presented, and finally the process fluctuation trajectory of pharmaceutical equipment is obtained.

[0048] Based on all nodes included in the fault propagation path of pharmaceutical equipment, each node is mapped as a segmentation identifier onto the process fluctuation trajectory of pharmaceutical equipment. The process fluctuation trajectory is segmented according to the mapping position of the node, so that each segment of the trajectory is uniquely associated with the corresponding node in the fault propagation path. Each segment of the trajectory retains its own fluctuation characteristics and time information, and finally the path fluctuation association segment of pharmaceutical equipment is obtained.

[0049] The fluctuation segments of each path of the pharmaceutical equipment are arranged in the order of the nodes of the fault propagation path. The numerical change trends of adjacent segments are compared and judged one by one. The consistency characteristics of each segment in terms of numerical rise and fall, fluctuation amplitude and change rate are analyzed. The trend judgment results of all segments are integrated to clarify the transmission change law of the fluctuation trend of process execution characteristics along the nodes of the fault propagation path. Finally, the deterioration transmission trend of pharmaceutical equipment along the fault propagation path is obtained.

[0050] Based on the equipment operation standards and health assessment specifications for pharmaceutical equipment, multiple health levels are set to correspond to different health conditions of the equipment. The deterioration trend of pharmaceutical equipment along the fault propagation path is matched with the judgment criteria of each health level. The corresponding health level is determined according to the degree of deterioration, the range of transmission, and the magnitude of change of the trend. The health level is then characterized and recorded in a standardized form to obtain the health status representation of the pharmaceutical equipment.

[0051] The beneficial effect is that by deeply linking the fluctuation analysis of process execution characteristics with the fault propagation path, a precise assessment of equipment health status from the perspective of process execution is achieved. First, the process fluctuation trajectory is extracted to restore the full picture of process characteristic changes. Then, node mapping and segmentation are used to make the trajectory correspond precisely with the fault propagation path. By judging the trend consistency, the deterioration pattern along the fault path is mined. Finally, a standardized health status representation is formed by health level calibration. This allows the assessment results of equipment health status to be closely integrated with the fault propagation mechanism, which not only improves the accuracy and pertinence of health status representation, but also provides a core basis for predicting the remaining service life of the equipment that is in line with the actual operating status of the equipment.

[0052] S5. Map the health status characteristics to the historical equipment service data of pharmaceutical equipment to predict the remaining service life of pharmaceutical equipment and obtain the risk time window of pharmaceutical equipment. In this embodiment of the invention, the step of mapping health status characteristics to historical equipment service data of pharmaceutical equipment to predict the remaining service life of pharmaceutical equipment and obtain the risk time window of pharmaceutical equipment includes: By jointly annotating the historical service data of pharmaceutical equipment, a sample library of historical healthy lifespan of pharmaceutical equipment is obtained. Centroid clustering was performed on the historical health status of the historical health life sample database to obtain a health status benchmark template for pharmaceutical equipment. Based on the health status benchmark template, the nearest neighbor search is performed on the health status representation to obtain the matching health status template of the pharmaceutical equipment. Based on the matching health status template, historical remaining life samples associated with the matching health status template are indexed from the historical health life sample library, and interval reduction is performed on the historical remaining life samples to obtain the remaining life prediction interval of pharmaceutical equipment. The remaining life prediction interval is defined as the risk time window for pharmaceutical equipment.

[0053] The process of performing nearest neighbor retrieval on the health status representation based on the health status benchmark template to obtain a matching health status template for the pharmaceutical equipment includes: A similarity space mapping is performed between the health status representation and the health status benchmark template to obtain the template difference sequence between the health status representation and the health status benchmark template; The minimum difference value in the template difference sequence is extracted as the nearest response value of the health status baseline template, and the nearest response value is marked with a position index to obtain the template index corresponding to the nearest response value. Use the health status benchmark template pointed to by the template index as the matching health status template for pharmaceutical equipment.

[0054] A comprehensive review of historical equipment service data for pharmaceutical manufacturing equipment was conducted. This data includes health status information, actual service life information, and various operating parameter information during the past operation of pharmaceutical equipment. For each set of historical equipment service data, its corresponding health status and remaining service life were marked one by one. At the same time, information related to the operating environment and usage conditions of the equipment was added. All the marked historical equipment service data were integrated and classified to form a complete and clearly marked historical health life sample library of pharmaceutical equipment.

[0055] All historical health status data in the historical health life sample database of pharmaceutical equipment are organized, and centroid clustering is carried out according to the characteristics and distribution patterns of the data. First, the core centroid of the cluster is determined, and then historical health status data with similar characteristics are grouped into the same cluster category. Each cluster category forms a standardized data template that can represent the health status characteristics of that category. After integrating the templates corresponding to all cluster categories, the health status benchmark template of pharmaceutical equipment is obtained.

[0056] The health status representation of pharmaceutical equipment and the health status benchmark template are placed in the same feature space for similarity space mapping. The feature differences of each template in the health status representation and the health status benchmark template are compared one by one. The feature differences obtained from each comparison are numerically represented. All numerical feature differences are arranged in the order of the corresponding templates to obtain the template difference sequence between the health status representation of pharmaceutical equipment and the health status benchmark template.

[0057] From the template difference sequence between the health status characterization of pharmaceutical equipment and the health status benchmark template, the smallest difference value is selected and determined as the nearest response value of the health status benchmark template. Based on the position of the nearest response value in the template difference sequence, a precise position index is marked, and the specific number of the health status benchmark template corresponding to the value is determined, thus obtaining the template index corresponding to the nearest response value.

[0058] Using the template index corresponding to the nearest response value as the reference, find the health status benchmark template of the pharmaceutical equipment corresponding to the index, and directly determine the template as the matching health status template of the pharmaceutical equipment.

[0059] Based on the matching health status template of pharmaceutical equipment, a comprehensive search is conducted in the historical health life sample database of pharmaceutical equipment to accurately index all historical remaining life samples that are associated with the matching health status template. This sample contains all data related to the remaining service life of the equipment corresponding to the matching health status template.

[0060] Data processing is performed on all historical remaining life samples indexed from the historical health life sample database. Abnormal data and extreme values ​​in the samples are removed. Then, the numerical range of the remaining valid historical remaining life samples is summarized and defined to determine the upper and lower limits of the sample values, forming a clear numerical range, and thus obtaining the remaining life prediction range of pharmaceutical equipment.

[0061] The remaining life prediction range of pharmaceutical equipment is standardized and calibrated. This range serves as the time range from the current operating state of the pharmaceutical equipment to the point where maintenance is required. It is directly determined as the risk time window of the pharmaceutical equipment, and the risk time window clearly presents the time boundary of the remaining life of the pharmaceutical equipment.

[0062] The beneficial effects are as follows: by accurately mapping and matching health status characteristics with historical equipment service data, a complete historical health life sample library is first constructed and a health status benchmark template is formed. Then, the current health status is accurately matched with the historical template through nearest neighbor retrieval. Based on the matching results, historical remaining life samples are indexed and the remaining life prediction interval is determined. Finally, it is marked as a risk time window, so that the prediction of the remaining service life of pharmaceutical equipment is based on historical actual service data, which greatly improves the accuracy and reliability of remaining life prediction. The defined risk time window also provides a clear and scientific time basis for the coordinated optimization of subsequent maintenance tasks and production plans, and effectively ensures the foresight of fault prediction.

[0063] S6. Based on the risk time window, the maintenance tasks and production plans of pharmaceutical equipment are optimized in a coordinated manner to obtain predictive maintenance work orders for pharmaceutical equipment.

[0064] In this embodiment of the invention, the step of co-optimizing the maintenance tasks and production plans of pharmaceutical equipment based on a risk time window to obtain predictive maintenance work orders for pharmaceutical equipment includes: By detecting the overlap between the risk time window and the production plan of pharmaceutical equipment, a set of conflicting production tasks for pharmaceutical equipment is obtained. Based on the conflict production task set, the urgency of the maintenance tasks to be performed on the pharmaceutical equipment is assigned to obtain a priority list of maintenance tasks for the pharmaceutical equipment. Based on the maintenance task priority list, the production periods in the production plan are rearranged in sequence to obtain an optimized production plan for pharmaceutical equipment. By integrating and arranging maintenance tasks from the optimized production plan and maintenance task priority list, predictive maintenance work orders for pharmaceutical equipment are obtained.

[0065] The risk time window of pharmaceutical equipment is compared segment by segment with the established production plan on the same time axis. The time range of the risk time window is checked to see if there is any overlap between the execution period of each production task in the production plan. All production tasks with time overlap are screened and collected one by one. These production tasks with time conflicts are integrated into a complete task set, and finally the conflict production task set of pharmaceutical equipment is obtained.

[0066] Based on the importance of production tasks, time overlap range, and potential production impact of failures involved in the conflict production task set of pharmaceutical equipment, the urgency of all pending maintenance tasks of pharmaceutical equipment is determined and assigned a value. According to the assignment results, the pending maintenance tasks are arranged in order from high to low. During the arrangement process, the execution requirements and time requirements of each maintenance task are clarified, and finally, a priority list of maintenance tasks for pharmaceutical equipment is obtained.

[0067] Based on the priority list of maintenance tasks for pharmaceutical equipment, and combined with the execution duration and time requirements of each maintenance task, the execution time periods of each production task in the original production plan are readjusted and sorted. Production tasks without conflicts are retained in their original time periods, while tasks within the conflicting production task set are postponed or advanced in time to ensure that the execution of production tasks and the implementation of maintenance tasks do not overlap in time, ultimately resulting in an optimized production plan for pharmaceutical equipment.

[0068] The optimized production plan and maintenance task priority list of pharmaceutical equipment are deeply integrated. According to the timeline, each maintenance task is precisely arranged in the idle time or designated time period of the optimized production plan. At the same time, the execution node of each maintenance task, the connection node of production tasks, and the execution requirements of each work are clearly defined. The integrated content is standardized and organized to form a standardized document that includes the time arrangement and task requirements of the entire production and maintenance process, and finally obtains the predictive maintenance work order of pharmaceutical equipment.

[0069] The beneficial effects are that by combining risk time windows with production plans for collaborative optimization, conflicting production tasks are first accurately identified, and maintenance tasks are then prioritized. Based on this, production plans are adjusted and maintenance tasks are integrated and arranged. The generated predictive maintenance work orders achieve precise matching between pharmaceutical equipment maintenance tasks and production plans in terms of time dimension. This ensures that high-priority maintenance tasks can be executed in a timely manner within the risk time window, avoiding production risks caused by equipment failures, while minimizing the impact of equipment maintenance on normal production, improving the utilization efficiency of production resources, and enabling efficient collaboration between equipment maintenance decisions and production operation planning, thus ensuring the continuity and stability of pharmaceutical production.

[0070] like Figure 2 The diagram shown is a functional block diagram of a pharmaceutical equipment fault diagnosis and prediction system based on artificial intelligence, provided in an embodiment of the present invention.

[0071] The AI-based pharmaceutical equipment fault diagnosis and prediction system 100 of this invention can be installed in an electronic device. Depending on the functions implemented, the AI-based pharmaceutical equipment fault diagnosis and prediction system 100 may include a multi-source data conditioning module 101, an adaptive blind source separation module 102, a fault mechanism tracing module 103, a health status assessment module 104, a remaining life prediction module 105, and a maintenance work order generation module 106. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0072] In this embodiment, the functions of each module / unit are as follows: The multi-source data conditioning module 101 is used to condition the multi-source operating data collected by the pharmaceutical equipment to obtain standardized time-series data of the pharmaceutical equipment. The multi-source operating data includes vibration signals, temperature data, pressure data and motor current data. The adaptive blind source separation module 102 is used to perform adaptive blind source separation on standardized time series data to obtain the physical state characteristics and process execution characteristics of pharmaceutical equipment. The fault mechanism tracing module 103 is used to trace the fault mechanism of pharmaceutical equipment based on the physical state characteristics, and obtain the fault propagation path of pharmaceutical equipment. The health status assessment module 104 is used to perform correlation assessment on process execution characteristics based on the fault propagation path to obtain the health status characterization of pharmaceutical equipment. The remaining life prediction module 105 is used to map the health status characterization to the historical equipment service data of the pharmaceutical equipment in order to predict the remaining life of the pharmaceutical equipment and obtain the risk time window of the pharmaceutical equipment. The maintenance work order generation module 106 is used to collaboratively optimize the maintenance tasks and production plans of pharmaceutical equipment based on risk time windows to obtain predictive maintenance work orders for pharmaceutical equipment.

[0073] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0074] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0075] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0076] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0077] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for fault diagnosis and prediction of pharmaceutical equipment based on artificial intelligence, characterized in that, The method includes: S1. Perform signal conditioning on the multi-source operating data collected from the pharmaceutical equipment to obtain standardized time-series data of the pharmaceutical equipment. The multi-source operating data includes vibration signals, temperature data, pressure data, and motor current data. S2. Adaptive blind source separation is performed on standardized time-series data to obtain the physical state characteristics and process execution characteristics of pharmaceutical equipment; S3. Based on the ontological state characteristics, trace the fault mechanism of pharmaceutical equipment to obtain the fault propagation path of pharmaceutical equipment. S4. Based on the fault propagation path, the process execution characteristics are correlated and evaluated to obtain the health status characterization of the pharmaceutical equipment. S5. Map the health status characteristics to the historical equipment service data of pharmaceutical equipment to predict the remaining service life of pharmaceutical equipment and obtain the risk time window of pharmaceutical equipment. S6. Based on the risk time window, the maintenance tasks and production plans of pharmaceutical equipment are optimized in a coordinated manner to obtain predictive maintenance work orders for pharmaceutical equipment.

2. The artificial intelligence-based fault diagnosis and prediction method for pharmaceutical equipment as described in claim 1, characterized in that, The process involves signal conditioning of the multi-source operational data collected from the pharmaceutical equipment to obtain standardized time-series data of the equipment. This multi-source operational data includes vibration signals, temperature data, pressure data, and motor current data. By collecting vibration signals, temperature data, pressure data, and motor current data of pharmaceutical equipment, multi-source operating data of the pharmaceutical equipment can be obtained. Sliding window autocorrelation analysis was performed on multi-source operational data to obtain the intra-source periodicity characteristics of pharmaceutical equipment; Cross-correlation mapping of periodic features within the source yields the time delay relationship of multi-source operational data; Based on the time delay relationship, phase compensation is performed on temperature data, pressure data, and motor current data to obtain synchronous multi-source data of pharmaceutical equipment. By reducing the intrinsic dimension of the synchronous multi-source data, the principal component feature matrix of pharmaceutical equipment is obtained. The principal component feature matrix is ​​normalized and registered to obtain standardized time-series data of pharmaceutical equipment.

3. The artificial intelligence-based fault diagnosis and prediction method for pharmaceutical equipment as described in claim 1, characterized in that, The adaptive blind source separation of standardized time-series data yields the physical state characteristics and process execution characteristics of the pharmaceutical equipment, including: The standardized time-series data were subjected to decorrelation processing to obtain the whitening data matrix of pharmaceutical equipment; The covariance matrix of the whitened data matrix is ​​orthogonally decomposed, and the decomposed eigenvalues ​​are arranged in descending order to obtain the eigenvalue spectrum sequence of pharmaceutical equipment. The eigenvalue spectrum sequence is ordered by successive drop ratios to obtain the source signal dimension of the pharmaceutical equipment. Based on the source signal dimension, source separation is performed on the whitened data matrix to obtain the independent source signal components of the pharmaceutical equipment; Power spectral density analysis is performed on independent source signal components to distinguish between high-frequency impact components and low-frequency trend components, and these components are used as the physical state characteristics and process execution characteristics of pharmaceutical equipment.

4. The artificial intelligence-based fault diagnosis and prediction method for pharmaceutical equipment as described in claim 3, characterized in that, The step of determining the order of the eigenvalue spectrum sequence by successive drop ratios to obtain the source signal dimension of the pharmaceutical equipment includes: The eigenvalues ​​in the eigenvalue spectrum sequence are paired up step by step to obtain the eigenvalue pairs of the eigenvalue spectrum sequence. The drop ratio of each successive eigenvalue pair is calculated to obtain the single-order drop ratio of the successive eigenvalue pairs; the formula for calculating the drop ratio is as follows: ; in, Indicates the first The single-level drop ratio of the steps, Indicates the first 1 eigenvalue, Indicates the order of descending order. 1 eigenvalue, Indicates the order of the eigenvalues. The range of values ​​is to , This represents the total number of eigenvalues ​​in the eigenvalue spectrum sequence; The single-order drop ratio is serialized and recombined to obtain the drop ratio sequence of the eigenvalue spectrum sequence; The order corresponding to the maximum value in the drop ratio sequence is taken as the dimension of the source signal of the pharmaceutical equipment.

5. The artificial intelligence-based fault diagnosis and prediction method for pharmaceutical equipment as described in claim 1, characterized in that, The method of tracing the fault mechanism of pharmaceutical equipment based on its ontological state characteristics to obtain the fault propagation path of the pharmaceutical equipment includes: Empirical wavelet decomposition is performed on the ontological state features to obtain the modal components of the pharmaceutical equipment. The information transmission direction and intensity between quantified modal components are used to construct an information flow matrix for pharmaceutical equipment. A fault information transmission network for pharmaceutical equipment is constructed using modal components as nodes and the transmission entropy value in the information flow matrix as the weight of the directed edges. Using the reciprocal of the transmission entropy as the path weight, the shortest path tracing is performed on the fault information transmission network to obtain the fault propagation path of the pharmaceutical equipment.

6. The artificial intelligence-based fault diagnosis and prediction method for pharmaceutical equipment as described in claim 1, characterized in that, The method of correlation evaluation of process execution characteristics based on fault propagation paths to obtain a health status characterization of pharmaceutical equipment includes: The process execution characteristics are extracted to obtain the process fluctuation trajectory of the pharmaceutical equipment; Based on the fault propagation path, node mapping and segmentation are performed on the process fluctuation trajectory to obtain the path fluctuation correlation segment of pharmaceutical equipment. By performing trend consistency judgment on the path fluctuation correlation segments, the deterioration transmission trend of pharmaceutical equipment along the fault propagation path can be obtained. By calibrating the health level of the degradation transmission trend, the health status of the pharmaceutical equipment can be characterized.

7. The artificial intelligence-based fault diagnosis and prediction method for pharmaceutical equipment as described in claim 1, characterized in that, The process of mapping health status characteristics to historical equipment service data of pharmaceutical equipment to predict the remaining service life of pharmaceutical equipment and obtain the risk time window of pharmaceutical equipment includes: By jointly annotating the historical service data of pharmaceutical equipment, a sample library of historical healthy lifespan of pharmaceutical equipment is obtained. Centroid clustering was performed on the historical health status of the historical health life sample database to obtain a health status benchmark template for pharmaceutical equipment. Based on the health status benchmark template, the nearest neighbor search is performed on the health status representation to obtain the matching health status template of the pharmaceutical equipment. Based on the matching health status template, historical remaining life samples associated with the matching health status template are indexed from the historical health life sample library, and interval reduction is performed on the historical remaining life samples to obtain the remaining life prediction interval of pharmaceutical equipment. The remaining life prediction interval is defined as the risk time window for pharmaceutical equipment.

8. The artificial intelligence-based fault diagnosis and prediction method for pharmaceutical equipment as described in claim 7, characterized in that, The process of performing nearest neighbor retrieval on the health status representation based on the health status benchmark template to obtain a matching health status template for the pharmaceutical equipment includes: A similarity space mapping is performed between the health status representation and the health status benchmark template to obtain the template difference sequence between the health status representation and the health status benchmark template; The minimum difference value in the template difference sequence is extracted as the nearest response value of the health status baseline template, and the nearest response value is marked with a position index to obtain the template index corresponding to the nearest response value. Use the health status benchmark template pointed to by the template index as the matching health status template for pharmaceutical equipment.

9. The artificial intelligence-based fault diagnosis and prediction method for pharmaceutical equipment as described in claim 1, characterized in that, The method of co-optimizing maintenance tasks and production plans for pharmaceutical equipment based on risk time windows to obtain predictive maintenance work orders for pharmaceutical equipment includes: By detecting the overlap between the risk time window and the production plan of pharmaceutical equipment, a set of conflicting production tasks for pharmaceutical equipment is obtained. Based on the conflict production task set, the urgency of the maintenance tasks to be performed on the pharmaceutical equipment is assigned to obtain a priority list of maintenance tasks for the pharmaceutical equipment. Based on the maintenance task priority list, the production periods in the production plan are rearranged in sequence to obtain an optimized production plan for pharmaceutical equipment. By integrating and arranging maintenance tasks from the optimized production plan and maintenance task priority list, predictive maintenance work orders for pharmaceutical equipment are obtained.

10. A pharmaceutical equipment fault diagnosis and prediction system based on artificial intelligence, characterized in that, The system is used to implement the artificial intelligence-based pharmaceutical equipment fault diagnosis and prediction method as described in claim 1, the system comprising: The multi-source data conditioning module is used to condition the multi-source operating data collected by the pharmaceutical equipment to obtain standardized time-series data of the pharmaceutical equipment. The multi-source operating data includes vibration signals, temperature data, pressure data and motor current data. The adaptive blind source separation module is used to perform adaptive blind source separation on standardized time series data to obtain the physical state characteristics and process execution characteristics of pharmaceutical equipment. The fault mechanism tracing module is used to trace the fault mechanism of pharmaceutical equipment based on the characteristics of the equipment's state, and to obtain the fault propagation path of the pharmaceutical equipment. The health status assessment module is used to perform correlation assessment of process execution characteristics based on fault propagation paths to obtain the health status characterization of pharmaceutical equipment. The remaining service life prediction module is used to map health status characteristics to historical equipment service data of pharmaceutical equipment in order to predict the remaining service life of pharmaceutical equipment and obtain the risk time window of pharmaceutical equipment. The maintenance work order generation module is used to collaboratively optimize the maintenance tasks and production plans of pharmaceutical equipment based on risk time windows, and obtain predictive maintenance work orders for pharmaceutical equipment.

Images