Flow control valve failure diagnosis method

Abstract

The application discloses a kind of flow control valve fault diagnosis methods, it is related to flow control valve fault diagnosis technical field, and its technical solution points are: establishing fault diagnosis system, the fault diagnosis system includes inference engine, interpreter, data acquisition module, flow prediction module, man-machine interaction module, database and knowledge base.The application provides a kind of new model updating method, avoids the situation that blind model updating leads to the breakdown of diagnostic system.The method can even be used in the case where the training data only includes one working condition, providing an effective model updating scheme;It also provides a fault diagnosis method under the condition of multiple working conditions and incomplete training samples, ensuring the performance of fault predictive alarm, while significantly reducing false alarm;In the offline model identification and online application stage, only DCS data, expert knowledge and user feedback information are used, which greatly saves the online cost of the system.

Description

Technical Field

[0001] This invention relates to the field of flow control valve fault diagnosis technology, and more specifically, to a method for diagnosing flow control valve faults. Background Technology

[0002] Flow control valves are indispensable actuators in industrial production, with even wider applications in the petrochemical industry. The frequent operation of valves and the corrosive nature of the media they handle lead to frequent valve failures. A malfunction in a control valve can paralyze the control circuit and even affect the entire production unit. Leaks of toxic, volatile, and flammable liquids threaten the safety of personnel and equipment. However, many failures are cumulative rather than sudden; for example, deformation caused by liquid impact on the valve body or wear on the valve stem. These valve failures are not observable under normal operating conditions. Except for regular maintenance, they are only discovered by operators when the failure escalates to the point of affecting production, leading to delayed detection. The large number of valves in complex industrial systems results in long manual maintenance cycles, high maintenance difficulty, and extremely high skill requirements for operators. Maintenance inevitably requires downtime and production stoppages, causing significant economic losses.

[0003] Therefore, some valve fault diagnosis systems utilizing DCS data modeling have been proposed. Fault diagnosis methods for flow control valves can establish valve models and achieve predictive diagnosis of valve faults through flow prediction, thereby avoiding blind valve maintenance, reducing the cost of routine factory maintenance, and improving system safety and reliability. However, complex industrial systems are characterized by multiple operating conditions and model drift. The data used to train the model is difficult to include multiple operating conditions; models built with training data containing only a few conditions are often inaccurate. Furthermore, a fixed model may mistake unknown normal operating conditions for faults, triggering false alarms and affecting the accuracy of fault diagnosis. However, blind model updates may lead to fault samples being used as training samples to update the model, causing the valve flow prediction model to collapse and become unusable. Summary of the Invention

[0004] The purpose of this invention is to solve the above-mentioned problems and provide a method for diagnosing faults in flow control valves.

[0005] The above-mentioned technical objective of the present invention is achieved through the following technical solution: a method for diagnosing faults in a flow control valve, comprising the following steps:

[0006] S1. Establish a fault diagnosis system, which includes an inference engine, an interpretation engine, a data acquisition module, a traffic prediction module, a human-computer interaction module, a database, and a knowledge base;

[0007] S2. In the offline stage, the parameters of the valve model are estimated using the parameter identification method, and the maximum and minimum values ​​of the valve opening OP in the training data, the standard deviation of the training error, and other information are stored. The alarm limit is determined based on this information.

[0008] S3. The data acquisition module sends the sample flow rate value PV and valve opening degree OP to the flow prediction module, and the flow rate setpoint value SP to the inference engine.

[0009] S4. The flow prediction module predicts the flow rate in real time based on the valve opening degree (OP). Calculate the valve failure score according to the given formula;

[0010] S5. The interpreter determines the fault type and fault level based on fault scoring and expert knowledge.

[0011] S6. Display the fault diagnosis information to the user through the human-computer interaction module and wait for the user's feedback.

[0012] S7. Based on the changes in the given flow value SP and user feedback, the inference engine determines whether to update the model and selects a specific plan for model updating.

[0013] S8. If remodeling is required, collect new training sample data and set the system to data collection state. At this time, use the mean prediction method to calculate the traffic prediction value. After collecting enough samples, build a new model and store the new model information and other data information needed in the process. If the inference engine determines that remodeling is not required, execute step S9.

[0014] S9. Wait for new samples, then proceed to step S3.

[0015] The present invention is further configured such that: in step S2, during the offline stage, the training data can be obtained from the DCS system, and the data selected is the data in the automatic control state of the control loop.

[0016] The present invention is further configured such that, in step S2, the alarm limit is determined as follows:

[0017] The OP and PV of the flow control valve satisfy the following relationship

[0018] PV = a × e b×OP (1)

[0019] Taking the logarithm of both sides simplifies to a linear model.

[0020] lnPV=lna+bOP (2)

[0021] Where a and b are the parameters that the flow valve model needs to identify. After training, the valve model parameters a and b are recorded and stored, and the modeling error for each training sample is calculated and recorded using the formula. The formula for calculating the error is expressed as follows:

[0022]

[0023] The modeling error E theoretically follows a Gaussian distribution with a mean of zero and a standard deviation of σ. Because different valves have varying service lives and manufacturing precision, a specific fault warning line needs to be set for each valve based on its model error. After obtaining the distribution of the modeling error, a significance level α can be given to determine the normal prediction error range. For a normal sample, the prediction error e should satisfy [the following condition].

[0024]

[0025] It can also be expressed as

[0026]

[0027] in These are the quantiles of the standard Gaussian distribution, which can be obtained by looking up a probability table. At this point, the parameters of the model and the fault warning line are determined.

[0028] In summary, the present invention has the following beneficial effects:

[0029] (1) This invention provides a novel model update method. This method uses expert knowledge to analyze real-time collected sample data and user feedback to determine the working conditions and whether the original model parameters need to be updated, thus avoiding the situation where blindly updating the model leads to the collapse of the diagnostic system. This method can even be used when the training data only includes one working condition, providing an effective model update solution.

[0030] (2) This invention provides a fault diagnosis method under multiple working conditions and incomplete training samples, which ensures the performance of fault prediction alarm, while significantly reducing the occurrence of false alarms, and further saves the factory's valve maintenance costs.

[0031] (3) In the offline model identification and online application stages, this invention only uses DCS data, expert knowledge and user feedback information, without the need to install hardware devices such as sensors, which greatly saves the system launch cost. Attached Figure Description

[0032] Figure 1 This is a system block diagram in an embodiment of the present invention;

[0033] Figure 2 This is a flowchart illustrating the implementation of the method in this embodiment of the invention;

[0034] Figure 3 This refers to the fault diagnosis performance in the absence of model updates in this embodiment of the invention.

[0035] Figure 4 This refers to the fault diagnosis performance in the case of model updates in the embodiments of the present invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be described in further detail below with reference to the embodiments and accompanying drawings. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0037] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the embodiments.

[0038] Example:

[0039] like Figures 1-4 As shown, a method for diagnosing faults in a flow control valve includes the following steps:

[0040] S1. Establish a fault diagnosis system, which includes an inference engine, an interpretation engine, a data acquisition module, a traffic prediction module, a human-computer interaction module, a database, and a knowledge base;

[0041] S2. In the offline stage, the parameters of the valve model are estimated using the parameter identification method, and the maximum and minimum values ​​of the valve opening OP in the training data, the standard deviation of the training error, and other information are stored. The alarm limit is determined based on this information.

[0042] S3. The data acquisition module sends the sample flow rate value PV and valve opening degree OP to the flow prediction module, and the flow rate setpoint value SP to the inference engine.

[0043] S4. The flow prediction module predicts the flow rate in real time based on the valve opening degree (OP). Calculate the valve failure score according to the given formula;

[0044] S5. The interpreter determines the fault type and fault level based on fault scoring and expert knowledge.

[0045] S6. Display the fault diagnosis information to the user through the human-computer interaction module and wait for the user's feedback.

[0046] S7. Based on the changes in the given flow value SP and user feedback, the inference engine determines whether to update the model and selects a specific plan for model updating.

[0047] S8. If remodeling is required, collect new training sample data and set the system to data collection state. At this time, use the mean prediction method to calculate the traffic prediction value. After collecting enough samples, build a new model and store the new model information and other data information needed in the process. If the inference engine determines that remodeling is not required, execute step S9.

[0048] S9. Wait for new samples, then proceed to step S3.

[0049] In this embodiment, the system is briefly described as follows: The entire fault diagnosis system includes an inference engine and an interpretation engine based on expert knowledge, a data acquisition module, a traffic prediction module, a human-computer interaction module, a database, and a knowledge base.

[0050] 1) Inference engine: Based on an expert knowledge base, it infers whether to update the model and which method to use to update the model based on changes in data or feedback from maintenance personnel.

[0051] 2) Explanation Unit: Converts the results of the flow prediction module from numbers into language that is easy for maintenance personnel to understand.

[0052] 3) Data acquisition module: Collects the required data through the DCS system.

[0053] 4) Flow prediction module: Identifies the parameters of the flow valve model and predicts the flow value in real time.

[0054] 5) Human-computer interaction module: Displays diagnostic results and collects feedback from maintenance personnel.

[0055] 6) Databases and knowledge bases: Store data and expert knowledge.

[0056] System block diagram as follows Figure 1 As shown.

[0057] Offline data selection

[0058] The data used in this invention to identify valve model parameters include valve opening degree (OP) and flow rate (PV). Training data can be obtained from a DCS system. To avoid the influence of operator intervention on the data, data should be selected when the control loop is in automatic control mode. To make the initial model as accurate as possible and reduce the time spent on offline modeling, the collected samples are sorted from largest to smallest according to OP values. Then, samples are selected at equal intervals based on experience as training data, assuming that N training sample data are obtained in the end. The range of OP values ​​is recorded [OP...]. min ,OP maxThe data is stored in the database and expert knowledge base, and this range is considered to be the normal operating opening range of the valve.

[0059] Offline model identification and alarm limit determination

[0060] The OP and PV of the flow control valve satisfy the following relationship:

[0061] PV = a × e b×OP (1)

[0062] Taking the logarithm of both sides simplifies to a linear model.

[0063] ln PV=lna+bOP (2)

[0064] Here, 'a' and 'b' are the parameters that the flow valve model needs to identify. These can be obtained using common algorithms, such as least squares optimization or particle swarm optimization. After training, the valve model parameters 'a' and 'b' are recorded and stored. The modeling error for each training sample is calculated and recorded using the formula: [Formula omitted for brevity].

[0065]

[0066] The modeling error E theoretically follows a Gaussian distribution with a mean of zero and a standard deviation of σ. Because different valves have varying service lives and manufacturing precision, a specific fault warning line needs to be set for each valve based on its model error. After obtaining the distribution of the modeling error, a significance level α can be given to determine the normal prediction error range. For a normal sample, the prediction error e should satisfy [the following condition].

[0067]

[0068] It can also be expressed as

[0069]

[0070] in These are the quantiles of the standard Gaussian distribution, which can be obtained by looking up a probability table. At this point, the parameters of the model and the fault warning line are determined.

[0071] Online application of the method

[0072] A. Fault scoring and interpretation machine description

[0073] The model identified in the offline phase can be used in the online phase, by collecting valve opening OP data in real time. new The predicted value of valve flow can be obtained. The prediction error for real-time samples is The valve's fault score is expressed as:

[0074]

[0075] The relationship between the fault score and fault level is pre-set in the interpreter, as detailed in Table 1.

[0076] Table 1 Fault Scoring Table

[0077]

[0078]

[0079] S in Table 1 i (i = 1, 2, 3) are set based on expert experience. When the fault score exceeds the normal range, i.e., there is a large deviation between the predicted and actual values, the cause of this phenomenon may be parameter changes caused by valve failure, model shift due to equipment aging, or significant external interference. In any case, triggering an alarm to remind staff to perform maintenance is reasonable. According to the scoring criteria in Table 1, the interpreter converts the score into a fault level that operators can understand, which is then displayed and pushed in real time through the human-machine interface.

[0080] B. Model Update and Inference Engine Description

[0081] Missing samples or normal operations such as switching operating conditions during modeling may cause the offline model parameters to become unsuitable for the current valve state, resulting in a large deviation between the predicted and actual flow rates and generating false alarms. Therefore, this invention designs an inference engine based on expert knowledge to determine the prerequisites and specific schemes for updating the model. Relevant descriptions are shown in Tables 2, 3, and 4. When the inference engine determines that the current valve needs a model update, the system begins to collect and store new samples as training data. This process is called the sample collection state and is displayed on the human-machine interface to prompt the user. In this state, the valve flow rate prediction value is obtained through the mean prediction method, that is, using the mean of the flow rates of the first m samples as the predicted flow rate value at the current moment, expressed by the formula:

[0082]

[0083] After the model is successfully updated, the new model parameters a and b, the standard deviation of the modeling error σ, and the maximum and minimum values ​​of OP are stored in the database.

[0084] Table 2 Model Update Scheme

[0085]

[0086]

[0087] Table 3 shows the model update schemes provided by the inference engine (based on the data).

[0088]

[0089] Note: For valve SP changes in a cascade secondary loop, SP change refers to a change greater than a fixed value. √ indicates the event occurred; × indicates the event did not occur; ○ indicates the occurrence of the event is irrelevant to the choice of solution.

[0090] Table 4 shows the model update schemes provided by the inference engine (based on user feedback).

[0091]

[0092] Experimental results

[0093] The experiment used data from a factory's flow control valve during normal operation to demonstrate the feasibility and effectiveness of the expert knowledge-based model update and fault diagnosis proposed in this invention. Following the method mentioned in the previous section, 500 samples were selected as training data and 2500 samples as test data. The relevant parameter settings mentioned above were: α = 0.1, m = 5, S1 = 20. The experimental results are as follows: Figure 3 and Figure 4 These represent the fault diagnosis performance with and without model updates, respectively. At the 1107th sample point, the change in SP was determined by the inference engine to be a change in operating condition, thus a model update was performed. Combined with... Figure 3 and Figure 4 As can be seen, the updated model provides better flow rate prediction and significantly reduces the false alarm rate. This experiment demonstrates the effectiveness and practical application value of the expert knowledge-based flow control valve fault diagnosis and model update method proposed in this invention.

[0094] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for diagnosing faults in a flow control valve, characterized in that, Includes the following steps: S1. Establish a fault diagnosis system, which includes an inference engine, an interpretation engine, a data acquisition module, a traffic prediction module, a human-computer interaction module, a database, and a knowledge base; S2. In the offline phase, the parameters of the valve model are estimated using parameter identification methods, and the valve opening values ​​from the training data are stored. The maximum and minimum values, and the standard deviation of the training error are used to determine the alarm limit; S3, The data acquisition module will collect the sample flow value. Valve opening The data is sent to the traffic prediction module, and the traffic setpoint is given. Send it into the inference engine; S4, the flow prediction module predicts the flow rate based on the valve opening. Real-time predicted flow values The valve failure score is calculated based on the given formula. S5. The interpreter determines the fault type and fault level based on fault scoring and expert knowledge. S6. Display the fault diagnosis information to the user through the human-computer interaction module and wait for the user's feedback. S7, Based on the given value of the flow rate Based on changes in data and user feedback, the inference engine determines whether to update the model and selects a specific update plan. S8. If remodeling is required, collect new training sample data and set the system to data collection state. At this time, use the mean prediction method to calculate the traffic prediction value. After collecting enough samples, build a new model and store the new model information and other data information needed in the process. If the inference engine determines that remodeling is not required, execute step S9. S9. Wait for new samples, then proceed to step S3.

2. The method for diagnosing faults in a flow control valve according to claim 1, characterized in that: In step S2, during the offline phase, the training data can be obtained from the DCS system, and the data selected is the data where the control loop is in automatic control mode.

3. The method for diagnosing faults in a flow control valve according to claim 1, characterized in that: In step S2, the alarm limit is determined as follows: Flow control valve and Satisfy the following relationship （1）； Taking the logarithm of both sides simplifies to a linear model. （2）； in and These are the parameters that the flow valve model needs to identify. After training, the valve model parameters are recorded and stored. , The modeling error for each training sample is calculated and recorded using a formula, which is expressed as follows: （3）； The modeling error E theoretically follows a mean of zero. The model uses a Gaussian distribution with standard deviation. Because different valves have varying service lives and manufacturing precision, a specific fault warning line needs to be set for each valve's model error. After obtaining the distribution of the modeling error, a significance level can be determined. Determine the normal prediction error range; for a normal sample prediction error... It should be satisfied; （4）； It can also be expressed as （5）； in , These are the quantiles of the standard Gaussian distribution, which can be obtained by looking up a probability table. At this point, the parameters of the model and the fault warning line are determined.

Images