A fault diagnosis method for abnormal naphtha quality parameters

Abstract

The application discloses a fault diagnosis method for abnormal naphtha quality parameters, which introduces an industrial reliability database to determine the prior fault probability of a root node, thereby solving the limitation of a traditional method which relies on subjective assignment of expert experience; and maps a fault tree into a Bayesian network to build a key importance index to quantify the contribution degree of each potential fault source to the abnormal naphtha quality parameters. When a fault warning is triggered, the method first calculates the posterior probability of each root node based on the Bayesian network; secondly, the key importance of each root node is calculated in combination with mechanism characteristics; and finally, the key importance is sorted to guide on-site personnel to preferentially check high-possibility nodes. The diagnosis method can effectively improve the fault checking efficiency of abnormal naphtha quality parameters.

Description

Technical Field

[0001] This invention relates to the field of oil refining and processing, specifically to a fault diagnosis method for abnormal naphtha quality parameters, used to guide on-site maintenance personnel to quickly pinpoint the root cause of the fault. Background Technology

[0002] Crude oil distillation is a core process in refining and chemical enterprises, separating crude oil into side-stream products such as naphtha, kerosene, and diesel. As the most important light oil product, naphtha's key quality parameters (such as final boiling point) are highly sensitive; any abnormalities will affect the product quality and yield of downstream units. With the increasing scale of refining and chemical units, system operation risks have intensified. Abnormal operating conditions such as salt buildup in trays and fluctuations in feed properties can easily cause naphtha quality parameters to deviate from normal ranges. Therefore, enterprises urgently need to conduct rapid fault diagnosis for abnormal naphtha quality parameters to enable timely remedial operations.

[0003] Fault trees, with their clear topological structure and intuitive causal relationships, have been widely used in industrial fault diagnosis. However, fault trees are limited to a qualitative analysis framework and cannot provide the probability of occurrence of each potential fault source. Field personnel can only check and verify each node along the fault tree, lacking priority guidance, which makes it difficult to improve diagnostic efficiency. To overcome these shortcomings, researchers generally map fault trees to Bayesian networks, using their bidirectional probabilistic reasoning capabilities to calculate the posterior probability of each root node, thereby achieving a quantitative ranking of potential fault sources.

[0004] However, in the diagnosis of abnormal naphtha quality parameters, when fault trees contain a large number of OR gate logic, the posterior probability ranking of each root node after mapping to a Bayesian network is basically consistent with the prior probability ranking, making it difficult to effectively distinguish the contribution degree of each potential fault source. Furthermore, the prior probabilities of the root nodes themselves largely rely on subjective assignments based on expert experience, lacking objective evidence and affecting the reliability of the diagnostic conclusions.

[0005] Based on the above, this invention proposes a fault diagnosis method for abnormal naphtha quality parameters. This method maps a fault tree to a Bayesian network, introduces an industrial reliability database to determine the prior failure probability of the root node, and constructs a key importance index to quantify the contribution of each potential fault source to the abnormal naphtha quality parameters. This guides on-site personnel to prioritize the investigation of high-probability fault nodes, improving fault diagnosis efficiency. Summary of the Invention

[0006] To address the aforementioned issues, this invention discloses a fault diagnosis method for abnormal naphtha quality parameters. The method aims to quantify the contribution of each potential fault source to the abnormal naphtha quality parameters and provide a priority for investigation, thereby guiding on-site maintenance personnel to quickly pinpoint the root cause of the fault.

[0007] This method includes the following steps:

[0008] (1) Based on the process mechanism and operation experience of crude oil distillation, determine the top event T, analyze the process variables that affect the top event T, and determine the intermediate events and basic events of the fault tree based on the process variables, sensor configuration and operation experience.

[0009] (2) For key process states in the fault tree that cannot be directly measured by sensors, they are estimated based on the mechanism model, and the calculation results are used as the mechanism model calculation events to supplement the fault tree;

[0010] (3) Analyze the logical relationships between events, and connect the top event, intermediate events, mechanism model calculation events and basic events through logic gates to establish a fault tree for abnormal naphtha quality parameters;

[0011] (4) Map the fault tree established in step (3) to the Bayesian network corresponding to the topology, wherein the top event, basic event, intermediate event and the mechanism model calculation event of the fault tree correspond to the leaf node, root node and intermediate node of the Bayesian network respectively, and the logic gates correspond to the directed edges and conditional probability distribution table between nodes.

[0012] (5) When the early warning system issues a fault early warning signal for the top event T, calculate the values ​​of each root node according to the following formula. The posterior probability of triggering the top event T:

[0013]

[0014] in, root node The prior probability of failure; For likelihood; denominator It is obtained by summing the joint distribution of the states of all root nodes in the network using the law of total probability.

[0015] (6) Calculate each root node based on the mechanism characteristics. criticality , to characterize its contribution to the top event T:

[0016]

[0017]

[0018] in, For the first The probability importance of each root node; For the first The critical importance of each root node; Let be the mechanism state indication function. When the estimated value of the key process state associated with the i-th root node by the mechanism model exceeds the set normal operating condition threshold, let... ,otherwise ; The mechanism excitation factor is determined based on the degree of deviation of the estimated values ​​of the key process states from the mechanism model.

[0019] (7) Based on the critical importance of each root node obtained in step (6), sort each root node in descending order and output the critical importance ranking result of each root node to indicate the high probability nodes to be investigated first.

[0020] In this method, each root node The prior fault probability is calculated using the following formula:

[0021]

[0022] Where t is the actual continuous operating time of the device since its most recent maintenance or replacement. The failure rate of the devices corresponding to each root node.

[0023] In this method, the failure rate is obtained from a process industry equipment reliability database. The database includes the Reliability Data Sheet for Marine and Onshore Equipment (OREDA), the Chemical Process Safety Center Process Equipment Reliability Database (CCPSPERD), the Instrument Reliability Data from the International Association of Automation, equipment failure statistics provided by equipment manufacturers, and an equipment failure rate statistics database established by enterprises based on long-term maintenance records.

[0024] In this method, likelihood Calculate using the following formula:

[0025]

[0026] in, This represents the total number of root nodes in the network. Representing the Random variables corresponding to each root node; This represents the specific binary state value of the node; root node In state The probability of; Given a specific combination of all root node states, the probability of the top event T is derived layer by layer from the conditional probability distribution table of each node in the network.

[0027] In this method, the mechanistic excitation factor Calculate using the following formula:

[0028]

[0029] in, This is the maximum incentive cap; This represents the deviation between the real-time estimated value of the critical process state associated with the i-th root node by the mechanistic model and the baseline value under normal operating conditions. and , respectively, represent the mean and standard deviation of the mechanism deviation under historical normal operating conditions; 𝑡𝑎𝑛ℎ(·) denotes the hyperbolic tangent function.

[0030] Beneficial effects:

[0031] This invention proposes a fault diagnosis method for abnormal naphtha quality parameters. First, addressing the issue of excessive reliance on subjective expert assignment for the prior probability of root nodes, an industrial reliability database is introduced to achieve objective calculation of the prior failure probability of root nodes. Simultaneously, a key importance index is constructed to quantify the contribution of each potential fault source to the abnormal naphtha quality parameters. Finally, based on the ranking of key importance, a troubleshooting priority is output to guide on-site maintenance personnel to prioritize high-probability nodes, quickly pinpoint the root cause of the fault, and improve fault diagnosis efficiency. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention.

[0033] Figure 2 It is a naphtha final boiling point fault tree.

[0034] Figure 3 It is a Bayesian network model of naphtha final boiling point. Detailed Implementation

[0035] The following detailed calculation process and specific operation procedures are provided with reference to the accompanying drawings and specific examples to further illustrate the present invention. This embodiment is implemented based on the technical solution of the present invention, but the scope of protection of the present invention is not limited to the following embodiment.

[0036] This case study uses an atmospheric and vacuum distillation unit in an oil refinery as an example. This unit processes 1000 t / h of crude oil. Due to fluctuations in crude oil properties, the unit experienced various malfunctions, including salt buildup in the atmospheric distillation column, leading to deterioration in separation accuracy and consequently causing abnormal changes in key quality parameters such as the final naphtha boiling point. Using a malfunction that occurred around June 1, 2024, as an example, the effectiveness of this method is verified, and its implementation process and diagnostic application results are explained in detail.

[0037] The implementation process of this case is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0038] (1) Analyze the process and instrumentation of the crude oil distillation atmospheric tower to determine the variables related to the abnormal final boiling point of naphtha. Through process analysis, the relevant variables include: atmospheric tower top temperature, tower top pressure, tower top cold reflux flow rate, atmospheric furnace outlet temperature, tray pressure drop, and reflux tank interface, etc.

[0039] (2) Taking the naphtha final boiling point anomaly as the top event, a fault tree is established based on existing sensors, operational experience, and process mechanisms, such as... Figure 2 As shown, detailed event descriptions are provided in Table 1.

[0040] Table 1. Fault Tree Event Description for Naphtha Final Distillation Point Anomalies

[0041]

[0042] (3) Map the established fault tree to the corresponding Bayesian network model, where the top event corresponds to the leaf node, each basic event corresponds to the root node, intermediate events and mechanism model calculation events correspond to intermediate nodes, and logic gates are transformed into directed edges and conditional probability distribution tables between nodes, establishing a network model as follows: Figure 3 The Bayesian network model of naphtha final boiling point is shown.

[0043] (4) Based on the industrial reliability database, check the failure rate of the equipment corresponding to each root node, and combine the actual continuous operating time of the device since the last maintenance. Calculate the prior failure probability of each root node according to the exponential distribution formula, and determine the initial probability distribution of each potential failure source.

[0044] (5) After the warning signal is triggered, the posterior probability of each root node under the condition of the top event is calculated according to Bayes' theorem. The calculation results of the prior and posterior probabilities of each root node are shown in Table 2. Under the all-OR gate logic structure, the posterior probability ranking of each root node is basically consistent with the prior probability. The posterior probability of the reflux pump failure BE12 with the highest prior failure rate is 22.89%, and the posterior probability of the actual failure source tower coiled salt BE111 is 17.27%.

[0045] Table 2 Prior and Posterior Probabilities of Each Root Node

[0046]

[0047] (6) Calculate the critical importance of each root node based on the mechanistic characteristics. The comparison between the simulated pressure drop value of the tray output by the mechanistic model and the measured value on site shows that the current pressure drop is consistently 15.2% higher than the baseline value under normal operating conditions, and the deviation shows a monotonically accumulating trend, which is consistent with the mechanistic characteristics of salt accumulation on the tray. Based on this, the mechanistic state indication function of BE111 is activated. After introducing the mechanistic characteristics, the calculation results of the critical importance of each root node are shown in Table 3. The critical importance of BE111 is 0.2717, which is the highest, and the diagnostic conclusion is consistent with the actual root cause.

[0048] Table 3 Calculation results of criticality of each root node

[0049]

[0050] In summary, this invention discloses a fault diagnosis method for abnormal naphtha quality parameters. By introducing an industrial reliability database, it achieves objective calculation of prior probabilities and constructs a key importance index that integrates mechanism characteristics to quantify the probability of occurrence of potential fault sources. This guides on-site personnel to prioritize the investigation of high-probability nodes and quickly pinpoint the root cause of the abnormal naphtha quality parameters.

Claims

1. A fault diagnosis method for abnormal naphtha quality parameters, characterized in that, The constructed fault tree is mapped to a Bayesian network corresponding to the topology, and key importance indicators are constructed to quantify the contribution of each potential fault source to the anomalies in naphtha quality parameters. This includes the following steps: (1) Based on the process mechanism and operation experience of crude oil distillation, determine the top event T, analyze the process variables that affect the top event T, and determine the intermediate events and basic events of the fault tree based on the process variables, sensor configuration and operation experience. (2) For key process states in the fault tree that cannot be directly measured by sensors, they are estimated based on the mechanism model, and the calculation results are used as the mechanism model calculation events to supplement the fault tree; (3) Analyze the logical relationships between events, and connect the top event, intermediate events, mechanism model calculation events and basic events through logic gates to establish a fault tree for abnormal naphtha quality parameters; (4) Map the fault tree established in step (3) to the Bayesian network corresponding to the topology, where the top event corresponds to the leaf node, each basic event corresponds to the root node, intermediate events and mechanism model calculation events correspond to intermediate nodes, and logic gates correspond to the directed edges and conditional probability distribution table between nodes. (5) When the early warning system issues a fault early warning signal for the top event T, calculate the values ​​of each root node according to the following formula. The posterior probability of triggering the top event T: in, root node The prior probability of failure; For likelihood; denominator It is obtained by summing the joint distribution of the states of all root nodes in the network using the law of total probability. (6) Calculate each root node based on the mechanism characteristics. criticality , to characterize its contribution to the top event T: in, For the first The probability importance of each root node; For the first The critical importance of each root node; Let be the mechanism state indication function. When the estimated value of the key process state associated with the i-th root node by the mechanism model exceeds the set normal operating condition threshold, let... ,otherwise ; The mechanism-inducing factor is determined based on the degree of deviation of the estimated values ​​of the key process states from the mechanism model. (7) Based on the critical importance of each root node obtained in step (6), sort each root node in descending order and output the critical importance ranking result of each root node to indicate the high probability nodes to be investigated first.

2. The fault diagnosis method for abnormal naphtha quality parameters according to claim 1, characterized in that, Root nodes are determined using an exponential distribution. Prior fault probability: Where t is the actual continuous operating time of the device since its most recent maintenance or replacement. The failure rate of the devices corresponding to each root node.

3. The fault diagnosis method for abnormal naphtha quality parameters according to claim 2, characterized in that, The failure rate is obtained from the process industry equipment reliability database. .

4. The fault diagnosis method for abnormal naphtha quality parameters according to claim 3, characterized in that, The process industry equipment reliability database includes the OREDA Reliability Data Manual for Marine and Onshore Equipment, the CCPS PERD Process Equipment Reliability Database from the Chemical Process Safety Center, the Instrument Reliability Data from the International Association of Automation, equipment failure statistics provided by equipment manufacturers, or equipment failure rate statistics databases established by enterprises based on long-term maintenance records.

5. The fault diagnosis method for abnormal naphtha quality parameters according to claim 1, characterized in that, Likelihood The calculation formula is as follows: in, This represents the total number of root nodes in the network. Representing the Random variables corresponding to each root node; This represents the specific binary state value of the node; root node In state The probability of; Given a specific combination of all root node states, the probability of the top event T is derived layer by layer from the conditional probability distribution table of each node in the network.

6. The fault diagnosis method for abnormal naphtha quality parameters according to claim 1, characterized in that, Mechanistic excitatory factors Calculate using the following formula: in, This is the maximum incentive cap; The deviation between the real-time estimated value of the critical process state associated with the i-th root node by the mechanism model and the baseline value under normal operating conditions; and , respectively, represent the mean and standard deviation of the mechanism deviation under historical normal operating conditions; 𝑡𝑎𝑛ℎ(·) denotes the hyperbolic tangent function.

Images