Reciprocating machine monitoring method based on multidimensional data analysis

Abstract

The invention relates to the technical field of multi-dimensional data analysis, in particular to a reciprocator monitoring method based on multi-dimensional data analysis, which comprises the following steps of: judging whether a prediction model is qualified or not based on a comprehensive fault deviation coefficient, and determining a reason for the disqualification of the prediction model based on an effective deviation difference value in allusion to the condition that the prediction model is disqualified; after the adjustment of the fault sample size is completed, judging that an unqualified reason is poor real-time performance of the prediction model based on the re-detected effective deviation difference value, and correcting the data block overlapping rate based on the task processing capability; based on the verification error rate, determining a reason for disqualification of the detection equipment; after the TCP window or the data acquisition frequency is adjusted, whether the prediction model is qualified or not is judged based on the adjusted fault deviation coefficient, and a model reconstruction notice is directly sent out for the unqualified prediction model. According to the invention, the fault prediction precision of the reciprocator is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of multidimensional data analysis technology, and in particular to a reciprocating machine monitoring method based on multidimensional data analysis. Background Technology

[0002] Domestic research on reciprocating machine fault prediction has gradually moved from the traditional signal analysis stage towards intelligent and multi-technology integration. However, the Ordos region still has significant shortcomings in the practical application of intelligent prediction technology. Currently, the overall fault diagnosis, condition monitoring, and prediction capabilities of reciprocating machines in the chemical industry remain at a low level, especially lacking efficient and reliable intelligent prediction systems. Therefore, how to effectively monitor the operating status of reciprocating machines in real time to provide early intervention for faults and achieve efficient monitoring of reciprocating machines has become a key technical problem that urgently needs to be solved by those skilled in the art.

[0003] Chinese Patent No. CN115977936A discloses a fault diagnosis system for reciprocating compressors, comprising: a data acquisition module, a diagnostic module, and an alarm module. The data acquisition module collects various operating parameters of the reciprocating compressor; the diagnostic module performs a comprehensive diagnosis of the compressor's status based on different types of operating parameters to determine whether a fault has occurred; when the diagnostic module detects a fault, the alarm module issues an alarm; the system also includes a judgment module, which first judges the validity of the collected operating parameters during fault diagnosis, then the diagnostic module diagnoses the status trends of the corresponding components based on the valid operating parameters, and finally performs fault diagnosis based on the valid operating parameters and status trends. This invention can eliminate misjudgments caused by invalid operating parameters, thereby improving the accuracy of diagnostic results. However, the above technical solution has the following problems: it does not consider that in complex actual situations, data acquisition and prediction models can affect the fault detection of the reciprocating compressor, leading to reduced accuracy in fault diagnosis based solely on operating parameters, thus affecting the efficiency and accuracy of subsequent adjustments. Summary of the Invention

[0004] To address this, the present invention provides a reciprocating machine monitoring method based on multidimensional data analysis, which overcomes the limitations of existing technologies that fail to consider the impact of data acquisition and prediction models on reciprocating machine fault detection under complex actual conditions. This results in reduced accuracy of fault diagnosis based solely on operating parameters, thereby affecting the efficiency and accuracy of subsequent adjustments.

[0005] To achieve the above objectives, the present invention provides a reciprocating machine monitoring method based on multidimensional data analysis, comprising: The comprehensive fault deviation coefficient is used to determine whether the prediction model is qualified. If the prediction model is unqualified, the reason for the unqualified prediction model is determined based on the effective deviation difference. The reason is that the prediction model is not accurate enough and the sample size of the fault type is determined based on the false negative rate. Alternatively, the detection equipment is abnormal and the reason for the unqualified detection equipment is determined based on the verification error rate. After the fault sample size is adjusted, the reason for non-compliance is determined based on the effective deviation difference of the re-detection, which is the poor real-time performance of the prediction model. The data block overlap rate is then corrected based on the task processing capability. The reasons for the failure of the detection equipment are determined based on the verification error rate: abnormal data transmission and TCP window determined based on bandwidth delay coefficient; or abnormal data acquisition and data acquisition frequency reduced based on the rate of change of the verification error rate. After adjustment, the model is judged to be qualified based on the adjusted fault deviation coefficient, and a model reconstruction notice is issued directly for unqualified model.

[0006] Furthermore, the process of determining whether the prediction model is qualified based on the comprehensive fault deviation coefficient includes: The detection sensors include sensors installed in the compressor and motor to detect acceleration, and sensors installed in the compressor to detect cylinder pressure. For each sensor, the fault probability output by the predictive model and the actual fault probability are obtained within a single monitoring cycle. The ratio of the absolute value of the difference between the fault probability output by the predictive model and the actual fault probability to the actual fault probability is recorded as the fault deviation coefficient of that sensor. It can be understood that the fault deviation coefficient includes the vibration fault deviation coefficient and the pressure fault deviation coefficient. The comprehensive fault deviation coefficient is determined based on the vibration fault deviation coefficient and the pressure fault deviation coefficient. The qualification of the prediction model is determined based on the comparison result between the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient. If the comprehensive fault deviation coefficient is less than or equal to the preset comprehensive fault deviation coefficient, the prediction model is deemed qualified. If the comprehensive fault deviation coefficient is greater than the preset comprehensive fault deviation coefficient, the reason for the failure of the prediction model is determined based on the effective deviation difference.

[0007] Furthermore, the process of determining the reasons for the failure of the prediction model based on the effective deviation difference includes: Obtain the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient within a single monitoring period, and record the difference between the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient as the effective deviation difference; If the effective deviation difference is greater than the preset effective deviation difference, the reason for non-compliance is determined to be low accuracy of the prediction model, and the sample size of the fault type is determined based on the false negative rate. If the effective deviation difference is less than or equal to the preset effective deviation difference, the reason for non-compliance is determined to be abnormality of the testing equipment, and the reason for non-compliance of the testing equipment is determined based on the verification error rate.

[0008] Furthermore, the process of determining the sample size of fault classes based on the false negative rate includes: The number of times a target device actually fails to detect a fault in historical data but the prediction model fails to issue a warning is recorded as the number of missed faults. The number of times a fault actually occurs and the model successfully issues a warning is recorded as the number of true positive faults. The sum of the number of missed faults and the number of true positive faults is recorded as the total number of actual faults. The ratio of the number of missed faults to the total number of actual faults is recorded as the missed fault rate. The number of fault samples is positively correlated with the false negative rate.

[0009] Furthermore, after the fault sample size is adjusted, the effective deviation difference is re-detected. If the effective deviation difference is greater than the preset effective deviation difference, the reason for non-compliance is determined to be poor real-time performance of the prediction model.

[0010] Furthermore, in response to the first preset condition, the process of correcting the data block overlap rate based on task processing capability includes: Obtain the remaining CPU corresponding to the task processing module within the prediction model, and record the remaining CPU as the task processing capacity. The data block overlap rate is corrected based on the task processing capability to reduce the data block overlap rate, and the correction magnitude of the data block overlap rate is negatively correlated with the task processing capability. The first preset condition is that the fault sample size adjustment is completed and the effective deviation difference is greater than the preset effective deviation difference.

[0011] Furthermore, the process of determining the reasons for the failure of testing equipment based on the verification error rate includes: Obtain the number of data packets with check errors and the total number of data packets received, and record the ratio of the number of data packets with check errors to the total number of data packets received as the check error rate; If the verification error rate is greater than the preset verification error rate, the reason for failure is determined to be abnormal data transmission, and the TCP window is determined based on the bandwidth delay coefficient; If the verification error rate is less than or equal to the preset verification error rate, the reason for non-compliance is determined to be abnormal data acquisition, and the data acquisition frequency is reduced based on the rate of change of the verification error rate.

[0012] Furthermore, in response to the second preset condition, the adjusted verification error rate is re-detected. If the verification error rate is greater than the preset verification error rate, it is determined whether to adjust the TCP window again based on the number of repeated adjustments. The process of determining whether to continue repeated adjustments based on the comparison results between the number of repeated adjustments and the preset threshold includes: If the number of repeated adjustments is less than the preset threshold, then it is determined that the adjustment will continue. If the number of repeated adjustments exceeds the preset threshold, the reason for failure is determined to be abnormal data acquisition; The second preset condition is that a single TCP window adjustment is completed.

[0013] Furthermore, in response to the third preset condition, the data acquisition frequency is adjusted based on the difference in the verification error rate, and the reduction magnitude is negatively correlated with the rate of change of the verification error rate; The third preset condition is that the TCP window completes repeated adjustments.

[0014] Furthermore, the process of determining whether the prediction model is qualified in response to the fourth preset condition includes: Reacquire the fault deviation coefficient; The fault deviation coefficient is compared with the preset fault deviation coefficient; If the fault deviation coefficient is less than or equal to the preset fault deviation coefficient, the prediction model is deemed qualified. If the fault deviation coefficient is greater than the preset fault deviation coefficient, the prediction model is deemed unqualified, and a model reconstruction notification is issued directly. The fourth preset condition is that the adjustment of the sample size for fault types is completed or the adjustment of the data acquisition frequency is completed.

[0015] Compared with the prior art, the beneficial effects of the present invention are that it determines whether the prediction model is qualified by comprehensively considering the fault deviation coefficient. In the case of unqualified prediction model, it determines the reason for the unqualified prediction model based on the effective deviation difference, making the judgment of the prediction model more accurate. At the same time, it re-determines the reason for the unqualified based on the adjusted effective deviation difference, thereby effectively eliminating the reasons for the unqualified prediction model, thus improving the efficiency of determining the reason for the unqualified, and thus improving the effectiveness of monitoring the reciprocating machine operation process.

[0016] Furthermore, this invention determines whether the prediction model is qualified based on the comparison result of the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient. The comprehensive fault deviation coefficient reflects the degree of fault deviation between the actual monitoring situation of each sensor in the compressor and motor and the fault probability output by the corresponding prediction model. This facilitates a comprehensive evaluation of the prediction model. The comprehensive deviation coefficient can quickly determine whether the prediction model is qualified, and when the prediction model is unqualified, the cause analysis can be performed quickly, thus effectively improving the efficiency of determining the cause of unqualification.

[0017] Furthermore, this invention determines the cause of the prediction model's failure based on the effective deviation difference, namely low prediction model accuracy or abnormal detection equipment. The effective deviation difference reflects the degree of deviation between the comprehensive deviation coefficient and the preset comprehensive deviation coefficient. Based on the effective deviation difference, the cause of the prediction model's failure can be analyzed more accurately, thereby improving the accuracy of the failure cause determination, and thus improving the prediction accuracy and stability of the reciprocating machine monitoring model, and improving the accuracy of the reciprocating machine fault prediction system.

[0018] Furthermore, this invention determines the fault sample size based on the false negative rate, which can more accurately determine the fault sample size based on the false negative rate, further ensuring the rationality of the fault sample size. While further realizing precise control of the fault sample size, it further improves the accuracy of the reciprocating machine fault prediction system.

[0019] Furthermore, after the fault sample size is adjusted, the present invention determines whether the prediction model is qualified based on the effective deviation difference detected again, and re-determines the reasons for non-compliance based on the determination results, so that the determination results are more in line with the actual operation of the reciprocating machine, avoiding the excessive adjustment of the data sample size caused by a single data adjustment method, thereby realizing the adaptive calibration process of the test system and further improving the accuracy of the reciprocating machine fault prediction system.

[0020] Furthermore, this invention corrects the data block overlap rate based on task processing capability. The data block overlap rate reflects the accuracy of the prediction system's data analysis and the smoothness of its output. By overlapping, it ensures that each data point is located in the central area of ​​a data block, thereby obtaining the most accurate analysis. This makes the determination of the data block overlap rate more reasonable and adaptable to the actual operating environment, further improving the accuracy of the reciprocating machine fault prediction system.

[0021] Furthermore, this invention determines the cause of equipment failure based on the verification error rate. The verification error rate reflects the probability of errors in the verification process, and the cause of equipment failure is determined to be abnormal data transmission or abnormal data acquisition. When data transmission is abnormal, the TCP window is determined based on the bandwidth delay coefficient. The bandwidth delay coefficient reflects the delay and bandwidth dynamics during actual data transmission, thereby determining the dynamic adjustment of the TCP window. This avoids the problem that a fixed TCP window cannot meet the actual application requirements, leading to idle transmission resources or transmission delays caused by not meeting the actual application requirements. When data acquisition is abnormal, the data acquisition frequency is adjusted to avoid the impact of data loss or data duplication caused by excessive data acquisition frequency, further improving the accuracy of the reciprocating machine fault prediction system.

[0022] Furthermore, this invention determines whether to continue repeatedly adjusting the TCP window based on the comparison result of the number of repeated adjustments and the preset threshold. After each adjustment is completed, the error rate after adjustment is re-checked to avoid invalid adjustments and the resulting waste of computing resources, thereby further improving the accuracy of the reciprocating machine fault prediction system.

[0023] Furthermore, this invention adjusts the data acquisition frequency based on the rate of change of the verification error rate, making the adjustment of the data acquisition frequency more in line with the actual situation of the application scenario. This avoids the problem of unreasonable adjustment of the data acquisition frequency in actual scenarios, and further improves the accuracy of the reciprocating machine fault prediction system while enabling rapid elimination of the causes of non-compliance. Attached Figure Description

[0024] Figure 1 This is a flowchart of the reciprocating machine monitoring method based on multidimensional data analysis according to the present invention; Figure 2 This is a flowchart illustrating the process of determining whether a prediction model is qualified based on the comparison result between the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient. Figure 3 This is a flowchart illustrating the process of determining the cause of non-compliance based on the effective deviation difference in this invention. Figure 4 This is a flowchart illustrating how the present invention determines the cause of non-compliance based on the verification error rate. Detailed Implementation

[0025] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0026] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0027] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0028] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0029] Please see Figures 1 to 4 As shown, this invention provides a reciprocating machine monitoring method based on multidimensional data analysis, comprising: The comprehensive fault deviation coefficient is used to determine whether the prediction model is qualified. If the prediction model is unqualified, the reason for the unqualified prediction model is determined based on the effective deviation difference. The reason is that the prediction model is not accurate enough and the sample size of the fault type is determined based on the false negative rate. Alternatively, the detection equipment is abnormal and the reason for the unqualified detection equipment is determined based on the verification error rate. After the fault sample size is adjusted, the reason for non-compliance is determined based on the effective deviation difference of the re-detection, which is the poor real-time performance of the prediction model. The data block overlap rate is then corrected based on the task processing capability. The reasons for the failure of the detection equipment are determined based on the verification error rate: abnormal data transmission and TCP window determined based on bandwidth delay coefficient; or abnormal data acquisition and data acquisition frequency reduced based on the rate of change of the verification error rate. After adjustment, the model is judged to be qualified based on the adjusted fault deviation coefficient, and a model reconstruction notice is issued directly for unqualified model.

[0030] Specifically, the process of determining whether the prediction model is qualified based on the comprehensive fault deviation coefficient includes: The detection sensors include sensors installed in the compressor and motor to detect acceleration, and sensors installed in the compressor to detect cylinder pressure. For each sensor, the fault probability output by the predictive model and the actual fault probability are obtained within a single monitoring cycle. The ratio of the absolute value of the difference between the fault probability output by the predictive model and the actual fault probability to the actual fault probability is recorded as the fault deviation coefficient of that sensor. It can be understood that the fault deviation coefficient includes the vibration fault deviation coefficient and the pressure fault deviation coefficient. The comprehensive fault deviation coefficient is determined based on the vibration fault deviation coefficient and the pressure fault deviation coefficient. The qualification of the prediction model is determined based on the comparison result between the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient. If the comprehensive fault deviation coefficient is less than or equal to the preset comprehensive fault deviation coefficient, the prediction model is deemed qualified. If the comprehensive fault deviation coefficient is greater than the preset comprehensive fault deviation coefficient, the reason for the failure of the prediction model is determined based on the effective deviation difference.

[0031] Specifically, in this embodiment, the comprehensive fault deviation coefficient = α1 × pressure fault deviation coefficient + α2 × vibration fault deviation coefficient; where α1 is the first weighting coefficient, α2 is the second weighting coefficient, and α1 + α2 = 1. The values ​​of α1 and α2 can be adaptively set by the user according to application requirements. It can be understood that the greater the influence of the pressure fault deviation coefficient on the comprehensive fault deviation coefficient, the larger the value of α1 and the smaller the value of α2. This invention provides a value of α1 and α2, where α1 = 0.5 and α2 = 0.5.

[0032] In this invention, the preset comprehensive fault deviation coefficient H0 = 0.13 is used. The comparison results between the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient are as follows: If the comprehensive fault deviation coefficient is less than or equal to the preset comprehensive fault deviation coefficient H0, the prediction model is deemed qualified. If the comprehensive fault deviation coefficient is greater than the preset comprehensive fault deviation coefficient H0, the reason for the failure of the prediction model is determined based on the effective deviation difference.

[0033] Specifically, the process of determining the reasons for the failure of the prediction model based on the effective deviation difference includes: Obtain the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient within a single monitoring period, and record the difference between the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient as the effective deviation difference; If the effective deviation difference is greater than the preset effective deviation difference, the reason for non-compliance is determined to be low accuracy of the prediction model, and the sample size of the fault type is determined based on the false negative rate. If the effective deviation difference is less than or equal to the preset effective deviation difference, the reason for non-compliance is determined to be abnormality of the testing equipment, and the reason for non-compliance of the testing equipment is determined based on the verification error rate.

[0034] Specifically, in this embodiment of the invention, the preset effective deviation difference value V0 = 0.05, and the comparison result between the effective deviation difference value and the preset effective deviation difference value is as follows: If the effective deviation difference is greater than the preset effective deviation difference V0, the reason for non-compliance is determined to be low accuracy of the prediction model, and the sample size of the fault type is determined based on the false negative rate. If the effective deviation difference is less than or equal to the preset effective deviation difference V0, the reason for non-compliance is determined to be abnormality of the testing equipment, and the reason for non-compliance of the testing equipment is determined based on the verification error rate.

[0035] Specifically, the process of determining the sample size of fault types based on the false negative rate includes: The number of times a target device actually fails to detect a fault in historical data but the prediction model fails to issue a warning is recorded as the number of missed faults. The number of times a fault actually occurs and the model successfully issues a warning is recorded as the number of true positive faults. The sum of the number of missed faults and the number of true positive faults is recorded as the total number of actual faults. The ratio of the number of missed faults to the total number of actual faults is recorded as the missed fault rate. The number of fault samples is positively correlated with the false negative rate.

[0036] Specifically, in this embodiment of the invention, if the false negative rate is greater than the second preset false negative rate F2, the number of fault samples is increased to 1.26 times the base number of fault samples. In this invention, the second preset false negative rate F2 = 10%. If the false negative rate is greater than the first preset false negative rate F1 and less than or equal to the second preset false negative rate, the number of fault samples is increased to 1.18 times the baseline number of fault samples. In this embodiment of the invention, the first preset false negative rate is 7%. If the false negative rate is less than or equal to the first preset false negative rate, the number of fault samples will be increased to 1.05 times the base number of fault samples. The value of the baseline fault sample size can be adaptively set by the user according to the actual application requirements. This invention provides a method for determining the value of the baseline fault sample size, which detects the corresponding fault sample size in the historical records that meet the user's requirements, filters out outliers, and records the average value of the fault sample size after removing outliers as the baseline fault sample size. The outlier filtering method can be, but is not limited to, the 3σ criterion method or the IQR method.

[0037] Specifically, after the fault sample size is adjusted, the effective deviation difference is re-detected. If the effective deviation difference is greater than the preset effective deviation difference, the reason for non-compliance is determined to be poor real-time performance of the prediction model.

[0038] Specifically, in response to the first preset condition, the process of correcting the data block overlap rate based on task processing capability includes: Obtain the remaining CPU corresponding to the task processing module within the prediction model, and record the remaining CPU as the task processing capacity. The data block overlap rate is corrected based on the task processing capability to reduce the data block overlap rate, and the correction magnitude of the data block overlap rate is negatively correlated with the task processing capability. The first preset condition is that the fault sample size adjustment is completed and the effective deviation difference is greater than the preset effective deviation difference.

[0039] Specifically, in this embodiment of the invention, if the task processing capability is greater than the second preset task processing capability G2, the data block overlap rate is reduced to 0.78 of the baseline data block overlap rate. In this invention, the second preset task processing capability G2 = 70%. If the task processing capacity is greater than the first preset task processing capacity G1 and less than or equal to the second preset task processing capacity, the data block overlap rate is reduced to 0.65 of the baseline data block overlap rate. In this embodiment of the invention, the first preset task processing capacity = 50%. If the task processing capacity is less than or equal to the first preset task processing capacity, the data block overlap rate is reduced to 0.56 of the baseline data block overlap rate. The user can adaptively set the value of the baseline data block overlap rate according to the actual application requirements. This invention provides a method for determining the value of the baseline data block overlap rate, which detects the data block overlap rate corresponding to the historical records that meet the user's requirements, filters out outliers, and records the average value of the data block overlap rate after removing outliers as the baseline data block overlap rate.

[0040] Specifically, the process of determining the reasons for the failure of testing equipment based on the verification error rate includes: Obtain the number of data packets with check errors and the total number of data packets received, and record the ratio of the number of data packets with check errors to the total number of data packets received as the check error rate; If the verification error rate is greater than the preset verification error rate, the reason for failure is determined to be abnormal data transmission, and the TCP window is determined based on the bandwidth delay coefficient; If the verification error rate is less than or equal to the preset verification error rate, the reason for non-compliance is determined to be abnormal data acquisition, and the data acquisition frequency is reduced based on the rate of change of the verification error rate.

[0041] Specifically, in this embodiment of the invention, the preset verification error rate K0 = 0.95, and the comparison result between the verification error rate and the preset verification error rate is as follows: If the verification error rate is greater than the preset verification error rate K0, the reason for failure is determined to be abnormal data transmission, and the TCP window is determined based on the bandwidth delay coefficient. If the verification error rate is less than or equal to the preset verification error rate K0, the reason for non-compliance is determined to be abnormal data acquisition, and the data acquisition frequency is reduced based on the rate of change of the verification error rate; wherein, the verification error rate difference = preset verification error rate - verification error rate; the bandwidth delay coefficient = bandwidth × round-trip time, wherein the round-trip time is the total time required for a single data packet to travel from the sending end to the receiving end and back to the sending end.

[0042] Specifically, in response to the second preset condition, the adjusted verification error rate is re-detected. If the verification error rate is greater than the preset verification error rate, it is determined whether to adjust the TCP window again based on the number of repeated adjustments. The process of determining whether to continue repeated adjustments based on the comparison results between the number of repeated adjustments and the preset threshold includes: If the number of repeated adjustments is less than the preset threshold, then it is determined that the adjustment will continue. If the number of repeated adjustments exceeds the preset threshold, the reason for failure is determined to be abnormal data acquisition; The second preset condition is that a single TCP window adjustment is completed.

[0043] Specifically, in this embodiment of the invention, the preset threshold N0 = 4 times, and the comparison result between the number of repeated adjustments and the preset threshold is as follows: If the number of repeated adjustments is less than the preset threshold N0, then it is determined that the adjustment should continue. If the number of repeated adjustments exceeds the preset threshold N0, the reason for non-compliance is determined to be abnormal data acquisition.

[0044] Specifically, in response to the third preset condition, the data acquisition frequency is adjusted based on the difference in the verification error rate, and the reduction rate is negatively correlated with the rate of change of the verification error rate. The third preset condition is that the TCP window completes repeated adjustments.

[0045] Specifically, in this embodiment of the invention, if the error rate difference is greater than the second preset error rate difference B2, the data acquisition frequency is reduced to 89% of the baseline data acquisition frequency. In this invention, the second preset error rate difference B2 = 5%. If the error rate difference is greater than the first preset error rate difference B1 and less than or equal to the second preset error rate difference, the data acquisition frequency is reduced to 92% of the baseline data acquisition frequency. In this invention, the first preset error rate difference B1 = 3%. If the error rate difference is less than or equal to the first preset error rate difference, the data acquisition frequency is reduced to 95% of the baseline data acquisition frequency. The value of the baseline data acquisition frequency can be adaptively set by the user according to the actual application requirements. This invention provides a method for determining the value of the baseline data acquisition frequency, which detects the corresponding data acquisition frequency in the historical records that meet the user's requirements, filters out outliers, and records the average value of the data acquisition frequency after removing outliers as the baseline data acquisition frequency.

[0046] Specifically, the process of determining whether the prediction model is qualified in response to the fourth preset condition includes: Reacquire the fault deviation coefficient; The fault deviation coefficient is compared with the preset fault deviation coefficient; If the fault deviation coefficient is less than or equal to the preset fault deviation coefficient, the prediction model is deemed qualified. If the fault deviation coefficient is greater than the preset fault deviation coefficient, the prediction model is deemed unqualified, and a model reconstruction notification is issued directly. The fourth preset condition is that the adjustment of the sample size for fault types is completed or the adjustment of the data acquisition frequency is completed.

[0047] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0048] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A reciprocating machine monitoring method based on multidimensional data analysis, characterized in that, include: The comprehensive fault deviation coefficient is used to determine whether the prediction model is qualified. If the prediction model is unqualified, the reason for the unqualified prediction model is determined based on the effective deviation difference. The reason is that the prediction model is not accurate enough and the sample size of the fault type is determined based on the false negative rate. Alternatively, the detection equipment is abnormal and the reason for the unqualified detection equipment is determined based on the verification error rate. After the fault sample size is adjusted, the reason for non-compliance is determined based on the effective deviation difference of the re-detection, which is the poor real-time performance of the prediction model. The data block overlap rate is then corrected based on the task processing capability. The reasons for the failure of the detection equipment are determined based on the verification error rate, namely abnormal data transmission and TCP window based on bandwidth delay coefficient, or abnormal data acquisition and reduction of data acquisition frequency based on the difference in verification error rate. After adjustment, the model is judged to be qualified based on the adjusted fault deviation coefficient, and a model reconstruction notice is issued directly for unqualified model prediction.

2. The reciprocating machine monitoring method based on multidimensional data analysis according to claim 1, characterized in that, The process of determining whether a prediction model is qualified based on the comprehensive fault deviation coefficient includes: The detection sensors include sensors installed in the compressor and motor to detect acceleration, and sensors installed in the compressor to detect cylinder pressure. For each sensor, the fault probability output by the predictive model and the actual fault probability are obtained within a single monitoring cycle. The ratio of the absolute value of the difference between the fault probability output by the predictive model and the actual fault probability to the actual fault probability is recorded as the fault deviation coefficient of the sensor. The fault deviation coefficient includes the vibration fault deviation coefficient and the pressure fault deviation coefficient. A comprehensive fault deviation coefficient is determined based on the vibration fault deviation coefficient and the pressure fault deviation coefficient. The qualification of the prediction model is determined based on the comparison result between the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient. If the comprehensive fault deviation coefficient is less than or equal to the preset comprehensive fault deviation coefficient, the prediction model is deemed qualified. If the comprehensive fault deviation coefficient is greater than the preset comprehensive fault deviation coefficient, the reason for the failure of the prediction model is determined based on the effective deviation difference.

3. The reciprocating machine monitoring method based on multidimensional data analysis according to claim 2, characterized in that, The process of determining the reasons for the failure of the prediction model based on the effective deviation difference includes: Obtain the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient within a single monitoring period, and record the difference between the comprehensive fault deviation coefficient and the preset comprehensive fault deviation coefficient as the effective deviation difference; If the effective deviation difference is greater than the preset effective deviation difference, the reason for non-compliance is determined to be low accuracy of the prediction model, and the sample size of the fault type is determined based on the false negative rate. If the effective deviation difference is less than or equal to the preset effective deviation difference, the reason for non-compliance is determined to be abnormality of the testing equipment, and the reason for non-compliance of the testing equipment is determined based on the verification error rate.

4. The reciprocating machine monitoring method based on multidimensional data analysis according to claim 3, characterized in that, The process of determining the sample size of fault classes based on the false negative rate includes: The number of times a target device actually fails to detect a fault in historical data but the prediction model fails to issue a warning is recorded as the number of missed faults. The number of times a fault actually occurs and the model successfully issues a warning is recorded as the number of true positive faults. The sum of the number of missed faults and the number of true positive faults is recorded as the total number of actual faults. The ratio of the number of missed faults to the total number of actual faults is recorded as the missed fault rate. The number of fault samples is positively correlated with the false negative rate.

5. The reciprocating machine monitoring method based on multidimensional data analysis according to claim 4, characterized in that, After the fault sample size is adjusted, the effective deviation difference is re-detected. If the effective deviation difference is greater than the preset effective deviation difference, the reason for non-compliance is determined to be poor real-time performance of the prediction model.

6. The reciprocating machine monitoring method based on multidimensional data analysis according to claim 5, characterized in that, In response to the first preset condition, the process of correcting the data block overlap rate based on task processing capability includes: Obtain the remaining CPU corresponding to the task processing module within the prediction model, and record the remaining CPU as the task processing capacity. The data block overlap rate is corrected based on the task processing capability to reduce the data block overlap rate, and the correction magnitude of the data block overlap rate is negatively correlated with the task processing capability. The first preset condition is that the fault sample size adjustment is completed and the effective deviation difference is greater than the preset effective deviation difference.

7. The reciprocating machine monitoring method based on multidimensional data analysis according to claim 3, characterized in that, The process of determining the reasons for the failure of testing equipment based on the verification error rate includes: Obtain the number of data packets with check errors and the total number of data packets received, and record the ratio of the number of data packets with check errors to the total number of data packets received as the check error rate; If the verification error rate is greater than the preset verification error rate, the reason for failure is determined to be abnormal data transmission, and the TCP window is determined based on the bandwidth delay coefficient; If the verification error rate is less than or equal to the preset verification error rate, the reason for non-compliance is determined to be abnormal data acquisition, and the data acquisition frequency is reduced based on the rate of change of the verification error rate.

8. The reciprocating machine monitoring method based on multidimensional data analysis according to claim 7, characterized in that, In response to the second preset condition, the adjusted verification error rate is re-detected. If the verification error rate is greater than the preset verification error rate, it is determined whether to adjust the TCP window again based on the number of repeated adjustments. The process of determining whether to continue repeated adjustments based on the comparison results between the number of repeated adjustments and the preset threshold includes: If the number of repeated adjustments is less than the preset threshold, then it is determined that the adjustment will continue. If the number of repeated adjustments exceeds the preset threshold, the reason for failure is determined to be abnormal data acquisition; The second preset condition is that a single TCP window adjustment is completed.

9. The reciprocating machine monitoring method based on multidimensional data analysis according to claim 8, characterized in that, In response to the third preset condition, the data acquisition frequency is adjusted based on the rate of change of the verification error rate, and the reduction is negatively correlated with the rate of change of the verification error rate. The third preset condition is that the TCP window completes repeated adjustments.

10. The reciprocating machine monitoring method based on multidimensional data analysis according to claim 9, characterized in that, The process of determining whether the prediction model is qualified in response to the fourth preset condition includes: Reacquire the fault deviation coefficient; The fault deviation coefficient is compared with the preset fault deviation coefficient; If the fault deviation coefficient is less than or equal to the preset fault deviation coefficient, the prediction model is deemed qualified. If the fault deviation coefficient is greater than the preset fault deviation coefficient, the prediction model is deemed unqualified, and a model reconstruction notification is issued directly. The fourth preset condition is that the adjustment of the sample size for fault types is completed or the adjustment of the data acquisition frequency is completed.

Images