Intelligent liquid chromatograph fault monitoring system

Abstract

The application discloses a kind of intelligent liquid chromatograph fault monitoring systems, it is related to liquid chromatograph fault monitoring technical field, comprising: data acquisition module, for the parameter data of liquid chromatograph is collected in real time by sensor;Operation verification module, for by collecting historical operation data, and mark corresponding failure or bad experimental result, establish operation-equipment association model, obtain operation mode influence coefficient;Hierarchical fault monitoring model establishment module, operation mode influence coefficient and the real-time data of sensor are used as model input, establish hierarchical fault monitoring model, output fault reminder;Fault early warning module, according to monitoring result generates real-time fault diagnosis report, indicates fault type.The application solves the problem that traditional liquid chromatograph fault monitoring relies on experience judgment and is easily disturbed by human factors, mistakenly attributes operation failure to equipment failure, or ignores potential failure.

Description

Technical Field

[0001] This invention relates to the field of liquid chromatograph fault monitoring technology, specifically to an intelligent liquid chromatograph fault monitoring system. Background Technology

[0002] Liquid chromatographs involve multiple subsystems, such as sample injection, pump control system, temperature control module, detector, and data processing module. The coupling effect between these modules makes the equipment highly susceptible to factors such as ambient temperature, unstable flow rate, and improper operation, leading to decreased detection accuracy and frequent malfunctions.

[0003] Currently, fault monitoring of liquid chromatographs largely relies on operator experience and regular maintenance. Traditional methods mainly focus on static threshold monitoring of single parameters. However, liquid chromatographs involve multiple sensors and components, and single threshold methods are insufficient to fully capture the complex dynamic relationships between different parts of the equipment, making it difficult to provide timely warnings in the early stages of a fault. This can lead to deviations in analytical data and even affect experimental results once a fault occurs. Furthermore, operator judgment based on experience is easily affected by environmental noise and human factors. It is difficult to address the issue that operational errors leading to misjudgments during fault prediction model training can affect the model's training input data, resulting in errors and hindering real-time, accurate fault warnings, thus impacting the accuracy of model outputs. Since minor anomalies that occur during continuous operation are often difficult to detect in a timely manner, delays in fault repair not only affect data reliability but may also cause instrument downtime and reduced production efficiency, leading to further misjudgments.

[0004] Therefore, the present invention provides an intelligent liquid chromatograph fault monitoring system. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent liquid chromatograph fault monitoring system to solve the existing problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent liquid chromatograph fault monitoring system, comprising:

[0007] The data acquisition module is used to acquire parameter data of the liquid chromatograph in real time through sensors;

[0008] The operation verification module is used to collect historical operation data and mark the corresponding faults or adverse experimental results to establish an operation-equipment correlation model and obtain the operation mode influence coefficient.

[0009] The graded fault monitoring model building module takes the influence coefficient of the operation mode and the real-time data of the sensor as the model input, builds a graded fault monitoring model, and outputs fault alerts;

[0010] The fault early warning module generates real-time fault diagnosis reports based on monitoring results, indicating the fault type.

[0011] A further improvement of this invention is that the operation verification module includes collecting historical sample injection volume and injection speed of operators as operation features, and collecting equipment failure type data corresponding to personnel operations when equipment malfunctions occur. The historical operation data and equipment failure type data are labeled to obtain a failure-operation label dataset. The time of operation is recorded and the corresponding equipment status is collected. The operation features and equipment status are combined to form a status-operation feature, which is used as training data for the operation-equipment association model. Based on the neural network model, the current operation behavior is used as input to calculate the SHAP of the current operation behavior, and the output is the influence gain of the current operation behavior on the probability of each failure occurrence, which is denoted as the operation mode influence coefficient of the failure type.

[0012] A further improvement of this invention is that the parameter data of the liquid chromatograph includes establishing a fault parameter dataset, standardizing and weighting the changes in equipment status corresponding to each fault type to obtain the fault result influence factor, marking the fault result influence factor, and sending it together with the corresponding fault type into the fault parameter dataset.

[0013] A further improvement of the present invention is that the graded fault monitoring model establishment module includes a rapid threshold screening unit, a time-series pattern analysis unit, and a deep fusion diagnosis unit;

[0014] The rapid threshold screening unit is used to automatically adjust the static threshold based on historical equipment data, output anomaly markers in the fault parameter dataset, and output a preliminary confidence level in combination with the operation mode influence coefficient of the fault type; when the preliminary confidence level is greater than 0.8, an alarm is triggered directly; when the preliminary confidence level is ∈ [0.4, 0.8] or the fault is combined with low fluctuation, the time series pattern analysis unit is triggered.

[0015] The time-series pattern analysis unit is used to set up a lightweight LSTM predictor, extract abnormal markers from the fast threshold screening unit, take equipment status data as input, predict the trend of equipment data change within the next 5 seconds, output potential fault types and distribution probabilities, when the maximum value of the distribution probability is greater than or equal to 75%, the corresponding fault type is output and an alarm is triggered, when the maximum value of the distribution probability is less than 75%, the deep fusion diagnostic unit is triggered.

[0016] The deep fusion diagnostic unit inherits data from the fast threshold screening unit and the time-series pattern analysis unit, introduces a deep learning model, and outputs accurate fault types.

[0017] A further improvement of this invention lies in that the rapid threshold screening unit is equipped with a confidence calculation strategy. This strategy calculates the mean μFa and standard deviation σFa of the impact factors of the corresponding fault results for any fault type over N historical time intervals t in the fault parameter dataset. This yields the dynamic threshold TFa for any fault type: TFa = μFa + k·σFa, where k represents the confidence interval coefficient. When the impact factor RFa corresponding to any current fault type is greater than the dynamic threshold corresponding to that fault type, it is identified as an anomaly, the fault type is output, and a preliminary confidence level is calculated. This represents the normalized influence coefficient of the operation mode for this fault type.

[0018] A further improvement of the present invention is that the fast threshold screening unit is also equipped with a low fluctuation data judgment strategy. The low fluctuation data judgment strategy obtains a low fluctuation judgment threshold TFa+Tnn by setting a fluctuation threshold Tnn. When there are more than or equal to 3 sets of data in the anomaly marker that are less than their corresponding low fluctuation judgment threshold, it is judged as a fault combined with low fluctuation, and the time sequence pattern analysis unit is triggered.

[0019] A further improvement of the present invention is that the time-series pattern analysis unit includes extracting device status x from the anomaly markers from the fast threshold screening unit. t As the primary input, and with the operational influence coefficient as an additional feature, a joint input is constructed. at this time This represents the operational influence coefficient for all fault types at time t. The LSTM model automatically learns the impact of these operational influence coefficients on changes in equipment state during time step calculations. During model execution, the current error is calculated. Add it to the input of the current time step, and in the hidden state h of the last layer. T Then add a raw score output layer to output the raw score z for each fault type. i =W z ·h T +b z W z Represents the weight matrix. b represents the current predicted value. z The bias term is then used to convert the original scores into a probability distribution, yielding the probability of each fault type.

[0020]

[0021] A further improvement of the present invention is that the deep fusion diagnostic unit is equipped with a dynamic fault prediction model. The dynamic fault prediction model is based on a convolutional neural network and extracts input data from the time-series pattern analysis unit as input. It includes an input layer, a processing layer, an operation impact prediction layer, an impact threshold checking layer, a dynamic control layer, and an output layer.

[0022] The input layer includes extracting the operation mode influence coefficients corresponding to all fault types and real-time equipment status data;

[0023] The processing layer processes the input data for each data channel using a fully connected layer, and then uses a concatenation layer to fuse and concatenate the input data into a vector X. de =Concatenate(vs',vo'), where Concatenate() represents the concatenation function, vs' represents the device state vector, and vo' represents the operation mode influence coefficient vector;

[0024] The operation impact prediction layer downsamples the convolutional features using convolutional and pooling layers, and predicts the fault impact coefficient (FIC) and operation impact coefficient (OIC) through a fully connected layer.

[0025] The predicted operation mode influence coefficient is checked by the influence threshold checking layer to see if it is greater than the set maximum value of operation influence, and then it is determined to truncate or send it to the dynamic control layer.

[0026] The dynamic control layer uses the operator behavior vector vo' as a compensation factor to dynamically adjust the processing path of the model, and then outputs it as the output layer, following the same method as the time-series pattern analysis unit to output the probability of the fault type.

[0027] A further improvement of the present invention is that the influence threshold checking layer includes setting a maximum value O for the influence of the operation. max If the predicted impact coefficient of the operation method is greater than the set maximum impact value, then truncation will be performed (OIC). adj =min(OIC,O max If the predicted operation mode influence coefficient is less than or equal to the set maximum operation influence value, it is input to the dynamic control layer.

[0028] A further improvement of this invention is that the dynamic control layer employs a gating mechanism combined with a compensation factor vo' to calculate the dynamic routing weight G = sig(vo'), and the feature after dynamic control is represented as F. route =GΘFIC+(1-G)ΘFIC', where Θ represents element-wise multiplication and FIC' represents the output without considering the operation behavior.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] 1. This invention first establishes an operation-equipment association model through an operation verification module. By adjusting the operation mode influence coefficient, the training data of the multi-level hierarchical fault monitoring model is adjusted, and the confidence level is adjusted in real time according to the operation behavior. Subsequently, the data error generated by the operation mode influence coefficient is used as training data, so that the model can "know" the potential interference of the current improper operation on the result when predicting the future state. In the deep fusion diagnosis unit, the operator behavior vector is used as a compensation factor to dynamically adjust the processing path of the model, adjust the processing strategy, reduce the risk of false alarms caused by operation, effectively weaken the interference of human operation on sensor data, reduce the risk of false alarms caused by operation errors, and improve the accuracy of fault diagnosis.

[0031] 2. A multi-level hierarchical fault monitoring model enables real-time monitoring of equipment status and can automatically adjust static and dynamic thresholds to ensure accurate identification of potential faults in the shortest possible time, save computing resources, and issue alarms in a timely manner. Attached Figure Description

[0032] Figure 1 This is a framework diagram of an intelligent liquid chromatograph fault monitoring system according to the present invention;

[0033] Figure 2 This is a flowchart of a hierarchical fault monitoring model for an intelligent liquid chromatograph fault monitoring system according to the present invention. Detailed Implementation

[0034] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0035] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0036] Example 1

[0037] Figure 1 This embodiment illustrates a framework diagram of an intelligent liquid chromatograph fault monitoring system, including:

[0038] The data acquisition module is used to collect parameter data of the liquid chromatograph in real time through sensors. Since each fault type of the liquid chromatograph is not determined by a single sensor parameter, the parameter data of the liquid chromatograph includes establishing a fault parameter dataset, standardizing and weighting the changes in equipment status corresponding to each fault type to obtain the fault result influence factor, marking the fault result influence factor, and sending it together with the corresponding fault type into the fault parameter dataset; that is, which parameters can affect the fault type.

[0039] The operation verification module is used to collect historical operation data and mark the corresponding faults or adverse experimental results.

[0040] Because the operating method affects the health status of the liquid chromatograph, but this effect is not absolute, and the magnitude of the error caused by operational errors is not critical, but it will affect the system's fault judgment, an operation-equipment correlation model is established to compensate for this effect and obtain the operation method influence coefficient. This includes collecting the operator's historical sample injection volume and injection speed as operation features, and collecting equipment fault type data corresponding to each operation. Fault types include pump seal leakage, proportional valve vibration, abnormal plunger rod friction, sensor drift, leakage, temperature abnormality, detector signal abnormality, autosampler position abnormality, etc. The historical operation data and equipment fault type data are labeled to obtain a fault-operation label dataset. The time of operation is recorded and the corresponding equipment status is collected. The equipment status includes pressure, temperature, flow rate, position and signal. The operation features and equipment status are combined to form a status-operation feature, which is used as training data for the operation-equipment correlation model. Based on the neural network model, the current operation behavior is used as input to calculate the SHAP of the current operation behavior. The output is the gain of the current operation behavior on the probability of occurrence of each fault, which is recorded as the operation method influence coefficient for that fault type.

[0041] By collecting historical sample injection volume and injection speed operation characteristics of operators and labeling them with corresponding equipment failure types, a failure-operation label dataset was constructed, enabling quantitative analysis of the risk contribution of operational behavior.

[0042] The graded fault monitoring model building module takes the influence coefficient of the operation mode and the real-time data of the sensor as the model input, builds a graded fault monitoring model, and outputs fault alerts;

[0043] The fault early warning module generates real-time fault diagnosis reports based on monitoring results, indicating the fault type.

[0044] Example 2

[0045] Figure 2This invention presents a flowchart of a graded fault monitoring model for an intelligent liquid chromatograph fault monitoring system. Based on the inventive concept of Embodiment 1, this invention provides a graded fault monitoring model, including:

[0046] Rapid threshold screening unit, time-series pattern analysis unit, and deep fusion diagnostic unit;

[0047] Although the error caused by a single operational mistake may be small, its cumulative effect can interfere with fault diagnosis. Therefore, by introducing the operational influence coefficient into the fault prediction model, the system can compensate for these interferences during the prediction process, thereby avoiding misjudgments caused by operational deviations. Specific implementation methods include:

[0048] The rapid threshold screening unit is used to automatically adjust the static threshold based on historical equipment data, output the abnormal markers in the fault parameter dataset, and output the preliminary confidence level in combination with the operation mode influence coefficient of the fault type.

[0049] The rapid threshold screening unit is equipped with a confidence calculation strategy. This strategy calculates the mean μFa and standard deviation σFa of the impact factors of the corresponding fault results for any fault type over N historical time intervals t in the fault parameter dataset. This yields the dynamic threshold TFa for any fault type: TFa = μFa + k·σFa, where k represents the confidence interval coefficient. When the impact factor RFa corresponding to any current fault type is greater than the dynamic threshold for that fault type, it is identified as an anomaly, the fault type is output, and a preliminary confidence score is calculated. α represents the normalized influence coefficient of the operation mode for this fault type. When the initial confidence level is greater than 0.8, an alarm is triggered directly.

[0050] The rapid threshold screening unit is also equipped with a low fluctuation data judgment strategy. The low fluctuation data judgment strategy obtains a low fluctuation judgment threshold TFa+Tnn by setting a fluctuation threshold Tnn. When there are more than or equal to 3 sets of data in the anomaly marker that are less than their corresponding low fluctuation judgment threshold, it is judged as a fault combined with low fluctuation, and the time series pattern analysis unit is triggered.

[0051] When the initial confidence level ∈ [0.4, 0.8] or the fault joint low fluctuation, the time series pattern analysis unit is triggered;

[0052] By combining sensor data with operational methods, the confidence level can be adjusted in real time based on operational behavior. When the impact is significant (operational behavior has a large influence): the confidence level will decrease, avoiding misjudging anomalies caused by operational errors as equipment malfunctions; when When the impact is relatively small (operational influence is weak): the confidence level is mainly determined by the equipment sensor data, that is, sensor abnormalities have a greater impact on the confidence level, reducing the risk of false alarms; enabling more accurate calculation of fault confidence, reducing false alarms, and improving diagnostic accuracy.

[0053] The time-series pattern analysis unit is used to set up a lightweight LSTM predictor, extract anomaly markers from the fast threshold screening unit, take equipment status data as input, predict the trend of equipment data changes within the next 5 seconds, and output potential fault types and distribution probabilities, including extracting equipment status x from the anomaly markers from the fast threshold screening unit. t As the primary input, and with the operational influence coefficient as an additional feature, a joint input is constructed. at this time This represents the operational influence coefficient for all fault types at time t. The LSTM model automatically learns the impact of these operational influence coefficients on changes in equipment state during time step calculations. During model execution, the current error is calculated. Add it to the input of the current time step, and in the hidden state h of the last layer. T Then add a raw score output layer to output the raw score z for each fault type. i =W z ·h T +b z W z Represents the weight matrix. b represents the current predicted value. z The bias term is then used to convert the original scores into a probability distribution, yielding the probability of each fault type.

[0054]

[0055] operational impact coefficient By combining it with equipment status data, the model can "know" the potential interference of current improper operation on the results when predicting future states. During the training process, the model will automatically learn how to adjust the impact on state prediction if the operation influence coefficient is high (e.g., improper operation), and thus make corresponding adjustments to the prediction output.

[0056] When the maximum probability of the distribution is greater than or equal to 75%, the corresponding fault type is output and an alarm is triggered. When the maximum probability of the distribution is less than 75%, it indicates that the probability of all faults is not high enough and more detailed analysis is needed to confirm the fault type. Therefore, the deep fusion diagnostic unit is triggered.

[0057] The deep fusion diagnostic unit is used to inherit the data from the fast threshold screening unit and the time-series pattern analysis unit, introduce a deep learning model, and carry a dynamic fault prediction model. The dynamic fault prediction model is based on a convolutional neural network and extracts the input data from the time-series pattern analysis unit as input. It includes an input layer, a processing layer, an operation impact prediction layer, an impact threshold checking layer, a dynamic control layer, and an output layer.

[0058] The input layer includes extracting the operation mode influence coefficients corresponding to all fault types and real-time equipment status data;

[0059] The processing layer processes the input data for each data channel using a fully connected layer, and then uses a concatenation layer to fuse and concatenate the input data into a vector X. de =Concatenate(vs',vo'), where Concatenate() represents the concatenation function, vs' represents the device state vector, and vo' represents the operation mode influence coefficient vector;

[0060] The operation impact prediction layer downsamples the convolutional features using convolutional and pooling layers, and predicts the fault impact coefficient (FIC) and operation impact coefficient (OIC) through a fully connected layer.

[0061] The influence threshold checking layer checks whether the predicted operation mode influence coefficient is greater than the set maximum operation influence value, and determines whether to truncate or send it to the dynamic control layer; the influence threshold checking layer includes setting the maximum operation influence value O. max If the predicted impact coefficient of the operation method is greater than the set maximum impact value, then truncation will be performed (OIC). adj =min(OIC,O max To prevent excessively large outliers from interfering with subsequent judgments, if the predicted operation mode influence coefficient is less than or equal to the set maximum operation influence value, it is input to the dynamic control layer.

[0062] The dynamic control layer uses the operator behavior vector vo' as a compensation factor to dynamically adjust the processing path of the model, and then outputs it as the output of the output layer, outputting the fault type probability in the same way as the time-series pattern analysis unit. The dynamic control layer uses a gating mechanism combined with the compensation factor vo' to calculate the dynamic routing weight G = sig(vo'), and the feature representation after dynamic control is F. route =GΘFIC+(1-G)ΘFIC', where Θ represents element-wise multiplication and FIC' represents the output without considering the operation behavior.

[0063] If G is large, it indicates significant operational intervention, and the model automatically reduces the risk of false alarms. In the dynamic control layer, the operator behavior vector is used as a compensation factor to dynamically adjust the model's processing path. When a significant impact of operational behavior on the equipment state is detected, the model can adjust its processing strategy to reduce the risk of false alarms caused by the operation. This dynamic control mechanism effectively distinguishes between short-term fluctuations caused by human operation and genuine equipment fault signals, reducing the false alarm rate and improving the accuracy of fault detection.

[0064] The threshold and weight settings can be based on the default settings of this invention, or they can be set by the operator.

[0065] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0066] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0067] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0068] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0069] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. An intelligent liquid chromatograph fault monitoring system, characterized in that: include: The data acquisition module is used to acquire parameter data of the liquid chromatograph in real time through sensors; The operation verification module is used to collect historical operation data and mark the corresponding faults or adverse experimental results to establish an operation-equipment correlation model and obtain the operation mode influence coefficient. The graded fault monitoring model building module takes the influence coefficient of the operation mode and the real-time data of the sensor as the model input, builds a graded fault monitoring model, and outputs fault alerts; The fault early warning module generates a real-time fault diagnosis report based on the monitoring results, indicating the fault type. The operation verification module includes collecting historical sample injection volume and injection speed of operators as operation features, and collecting equipment failure type data corresponding to personnel operations when equipment malfunctions. The historical operation data and equipment failure type data are labeled to obtain a failure-operation label dataset. The time of operation is recorded and the corresponding equipment status is collected. The operation features and equipment status are combined to form a status-operation feature, which is used as training data for the operation-equipment association model. Based on the neural network model, the current operation behavior is used as input to calculate the SHAP of the current operation behavior and output the impact gain of the current operation behavior on the probability of each failure occurrence, which is recorded as the operation mode influence coefficient of the failure type. The hierarchical fault monitoring model establishment module includes a rapid threshold screening unit, a time-series pattern analysis unit, and a deep fusion diagnosis unit. The rapid threshold screening unit is used to automatically adjust the static threshold based on historical equipment data, output anomaly markers in the fault parameter dataset, and output a preliminary confidence level in combination with the operation mode influence coefficient of the fault type; when the preliminary confidence level is greater than 0.8, an alarm is triggered directly; when the preliminary confidence level is ∈ [0.4, 0.8] or the fault is combined with low fluctuation, the time series pattern analysis unit is triggered. The time-series pattern analysis unit is used to set up a lightweight LSTM predictor, extract abnormal markers from the fast threshold screening unit, take equipment status data as input, predict the trend of equipment data change within the next 5 seconds, output potential fault types and distribution probabilities, when the maximum value of the distribution probability is greater than or equal to 75%, the corresponding fault type is output and an alarm is triggered, when the maximum value of the distribution probability is less than 75%, the deep fusion diagnostic unit is triggered. The deep fusion diagnostic unit inherits data from the fast threshold screening unit and the time-series pattern analysis unit, introduces a deep learning model, and outputs accurate fault types.

2. The intelligent liquid chromatograph fault monitoring system according to claim 1, characterized in that: The parameter data of the liquid chromatograph includes establishing a fault parameter dataset, standardizing and weighting the changes in equipment status corresponding to each fault type to obtain the fault result influence factor, marking the fault result influence factor, and sending it along with the corresponding fault type into the fault parameter dataset.

3. The intelligent liquid chromatograph fault monitoring system according to claim 1, characterized in that: The rapid threshold screening unit is equipped with a confidence calculation strategy, which calculates the mean of the influence factor of the corresponding fault result for any fault type in the fault parameter dataset based on N historical time intervals t. and standard deviation To obtain the dynamic threshold for any fault type 'k' represents the confidence interval coefficient. When the result impact factor RFa corresponding to any fault type is greater than the dynamic threshold corresponding to that fault type, it is judged as an anomaly, the fault type is output, and the preliminary confidence level is calculated. , This represents the normalized influence coefficient of the operation mode for this fault type.

4. The intelligent liquid chromatograph fault monitoring system according to claim 3, characterized in that: The rapid threshold screening unit is also equipped with a low fluctuation data judgment strategy. The low fluctuation data judgment strategy obtains the low fluctuation judgment threshold TFa+Tnn by setting a fluctuation threshold Tnn. When there are more than or equal to 3 sets of data in the anomaly marker that are less than their corresponding low fluctuation judgment threshold, it is judged as a fault combined with low fluctuation, and the time sequence pattern analysis unit is triggered.

5. The intelligent liquid chromatograph fault monitoring system according to claim 1, characterized in that: The time-series pattern analysis unit includes extracting device status from anomaly markers from the fast threshold screening unit. As the primary input, and with the operational influence coefficient as an additional feature, a joint input is constructed. ,at this time This represents the operational influence coefficient for all fault types at time t. The LSTM model automatically learns the impact of these operational influence coefficients on changes in equipment state during time step calculations. During model execution, the current error is calculated. Add it to the input of the current time step, and in the hidden state of the last layer. Then add a raw score output layer to output the raw scores for each fault type. , Represents the weight matrix. This represents the current predicted value. The bias term is then used to convert the original scores into a probability distribution, yielding the probability of each fault type. , .

6. The intelligent liquid chromatograph fault monitoring system according to claim 1, characterized in that: The deep fusion diagnostic unit is equipped with a dynamic fault prediction model. The dynamic fault prediction model is based on a convolutional neural network and extracts input data from the time-series pattern analysis unit as input. It includes an input layer, a processing layer, an operation impact prediction layer, an impact threshold checking layer, a dynamic control layer, and an output layer. The input layer includes extracting the operation mode influence coefficients corresponding to all fault types and real-time equipment status data; The processing layer processes the input data for each data channel using a fully connected layer, and then uses a concatenation layer to fuse and concatenate the input data into a vector. , This represents the concatenation function. Represents the device state vector. This represents the vector of coefficients affected by the operation method; The operation impact prediction layer downsamples the convolutional features using convolutional and pooling layers, and predicts the fault impact coefficient (FIC) and operation impact coefficient (OIC) through a fully connected layer. The predicted operation mode influence coefficient is checked by the influence threshold checking layer to see if it is greater than the set maximum value of operation influence, and then it is determined to truncate or send it to the dynamic control layer. The dynamic control layer utilizes operator behavior vectors As a compensation factor, the processing path of the dynamically adjusted model is then output by the output layer, which outputs the fault type probability in the same way as the time-series pattern analysis unit.

7. The intelligent liquid chromatograph fault monitoring system according to claim 6, characterized in that: The impact threshold checking layer includes setting a maximum value for the impact of an operation. If the predicted impact coefficient of the operation method is greater than the set maximum impact value, then truncation will be performed. If the predicted operation mode influence coefficient is less than or equal to the set maximum operation influence value, it is input to the dynamic control layer.

8. The intelligent liquid chromatograph fault monitoring system according to claim 6, characterized in that: The dynamic control layer employs a gating mechanism combined with compensation factors. Calculate dynamic routing weights The characteristics after dynamic adjustment are expressed as follows: ,in This represents element-wise multiplication. This indicates the output without considering the operation behavior.

Images