Probability-driven pump unit system recession identification method

Abstract

The invention discloses a probability-driven pump unit system recession identification method, and relates to the technical field of pump unit system state monitoring, and the method comprises an offline stage and an online stage; the off-line stage comprises the following steps: S1, carrying out a vibration signal acquisition test on the pump unit; s2, performing signal processing on a plurality of groups of original vibration signals acquired in unit time to obtain a plurality of groups of sub-band discrete signals, and storing the sub-band discrete signals as to-be-processed signals; according to the method, based on the hidden Markov model for a period of time, the faults of the pump unit system are distinguished in the form of the cumulative product of the probability within a period of time instead of depending on the state at a certain moment, the influence of random errors on the system is effectively avoided, misjudgment is reduced, the system state recognition rate and accuracy are improved, and potential accident risks are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of state monitoring of a pump unit system, and in particular to a probability-driven pump unit system degradation identification method. Background Art

[0002] In recent years, with the development of intelligent remote technology, people have gradually become accustomed to judging the health or decline of pump units by monitoring the characteristic signals of pumps (such as vibration signals), which has facilitated the remote management and maintenance of pump units. The traditional cognitive logic is to directly obtain the current operating signal of the unit, derive the current operating status of the pump unit through simple analysis, and directly equate this status with the current health status of the unit. The above scheme mainly uses the corresponding algorithm to extract characteristic parameters from the monitoring signal, and then compares them with several standard features that have been calibrated in the historical database. The state represented by the standard feature with the closest numerical feature is regarded as the current operating state of the pump.

[0003] However, during actual operation of a pump unit, due to the gradual aging of certain components, the pump's health status may exhibit multiple operating states with a certain probability over a period of time. These operating states may be normal or abnormal. However, only when a component of the pump unit has aged to a certain extent, causing the pump unit to consistently operate abnormally, can the pump unit be considered to be in a declining (unhealthy) state and require maintenance.

[0004] In these situations, existing health diagnostic methods for pump units are clearly no longer applicable. If a healthy pump unit exhibits multiple abnormal operating states over a period of time, existing monitoring and diagnostic methods may misdiagnose the pump unit as unhealthy. Similarly, a degraded pump unit may exhibit multiple normal operating states over a period of time. Existing fault diagnosis methods may misdiagnose the degraded pump unit as healthy, thus creating potential accident risks.

[0005] Based on this, a probability-driven pump unit system degradation identification method is now provided, which can eliminate the drawbacks of existing devices. Summary of the Invention

[0006] The object of the present invention is to provide a probability-driven pump unit system degradation identification method to solve the problem of the shortcomings of modern monitoring methods in the background art that are prone to misjudgment.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A probability-driven pump unit system degradation identification method includes offline and online stages;

[0009] The offline phase includes the following steps:

[0010] S1: Conduct vibration signal acquisition test on the pump unit;

[0011] S2: performing signal processing on a plurality of groups of original vibration signals collected within a unit time to obtain a plurality of groups of sub-band discrete signals, and saving them as signals to be processed;

[0012] S3: extracting features from the collected signals of several groups of sub-band discrete signals containing several discrete data points to obtain three original feature parameter values;

[0013] S4: The three original characteristic parameter values ​​collected are normalized using the maximum value method to obtain three normalized value points;

[0014] S5: constructing three-dimensional space points based on the one-to-one correspondence between the three normalized numerical points and the state, and dividing the collected three-dimensional space points into two categories, representing the normal operation state and abnormal operation state of the pump, respectively. Then, the values ​​of the data center points of the two categories are found to obtain the cluster center points;

[0015] S6: Based on the distance between the normalized numerical point and the cluster center, the target's current operating status is distinguished, and the normal operation state and abnormal operation state of the pump are automatically classified to complete the construction of the offline stage data set.

[0016] On the basis of the above technical solutions, the present invention also provides the following optional technical solutions:

[0017] In an optional solution, the online stage includes the following steps:

[0018] S7: Install a vibration sensor on the pump to be tested, measure the vibration acceleration signal of the pump over a period of time, and perform signal processing, feature extraction, and normalization operations on the obtained raw signal in the same manner as steps S1-S4 to obtain several groups of numerical points. These groups of numerical points are compared with the cluster center points of the normal operation state and abnormal operation state of the pump trained in step S5;

[0019] S8: Collect health and decay status probability data for the same type of pumps, and calculate the initial state probability P1, pressure change probability P2 caused by the health status of the pump, and normal and abnormal operation probabilities P3 caused by pressure state changes of the same type of pumps, and integrate the data into a hidden Markov model λ = (A, B, π);

[0020] S9: Construct a set of hidden states and observable states, Define Q as the set of all possible hidden states, V as the set of all possible observed states, N as the number of possible hidden states, and M as the number of all possible observable states;

[0021] S10: Construct hidden state and observable state sequence, A sequence of length T, I is the corresponding hidden state sequence, O is the corresponding observable sequence, any hidden state i t ∈Q, any observation state o t ∈V;

[0022] S11: Construct hidden state probability matrix:

[0023] The hidden state at time t is i t =q i , the hidden state at time t+1 is i t+1 =q j , the state transition probability a from time t to time t+1 ij It can be expressed as: ai j =P(i t+1 =q j |i t =q i );

[0024] a ij The hidden state Markov state transition matrix A can be formed: A=[a ij ] N×N ;

[0025] S12: Construct observable state probability matrix:

[0026] The hidden state at time t is i t =q j , the corresponding observation state is o t =v k , the state v at this moment k In the hidden state q j The probability of generating is b j (k), satisfies: b i (k)=P(o t =v k |i t =q j );

[0027] b j (k) can form the probability matrix B of the observed state generation: B = [b j (k)] N×M

[0028] S13: Construct the initial state probability matrix:

[0029] The hidden state probability distribution π at the initial time t=1 is defined as: π=[π(i)] N , where π(i)=P(i1=q i );

[0030] S14: Calculate the probability of unobservable state:

[0031] For a given observation sequence O, the probability of an unobservable feature it appearing at time t is calculated as: P t (i t |λ)=πA t-1 ;

[0032] S15: Calculate the probability of observable state:

[0033] Calculate the probability of a certain observable state Ot occurring at time t as P t (O t |λ)=πA t-1 B;

[0034] S16: Calculate the state sequence probability:

[0035] At time t, for a given observation sequence O, the probability of occurrence of the state sequence is

[0036] S17: Perform state probability comparison, record the corresponding vibration signals of the pump within a period of time t, calculate and identify the operating state of the pump at each moment in the time period in step S7, and obtain the state sequence of the pump unit within the period of time t based on the identified operating state;

[0037] Assuming that the pump is in a healthy state and a decaying state, calculate the probability Pt(O|λh) of the state sequence when the pump unit is in a healthy state and the probability Pt(O|λf) of the state sequence when the pump unit is in a decaying state according to steps S10-S16, and compare the value of Pt(O|λh) with Pt(O|λf). If P t (O|λ h )>P t (O|λ d ) the pump unit under test is judged to be in a healthy state; otherwise, the pump unit is judged to be in a decayed state.

[0038] In an optional solution: in step S2, assume that the number of groups of original vibration signals collected within the unit time t is s(t), use the wavelet packet formula to process the signal, select Bior3.7 wavelet as the mother wave function, ψ * (·) is the conjugate function of ψ(·), according to the formula Process the original signal to obtain a new signal S(j,k);

[0039] Where j and k represent the scale parameter and translation parameter respectively; ψ * (·) represents the size of the mother wave ψ(·) and the transformation mother wave function; j and k represent the scale parameter and translation parameter respectively; the formula is used. The 12th group of sub-band signals S (j=9, k=12) of the 9th layer are saved as filtered signals to be processed.

[0040] In an optional solution: in step S3: the collected multiple groups of sub-band discrete signals and the signal Sn containing m discrete data points are processed The three formulas are used to obtain the absolute mean S abs , root mean square S rms , variance S δ These three characteristic parameters are used as original characteristic parameter values.

[0041] In an optional solution: in step S4, the three original characteristic parameter values ​​collected are The formula uses the maximum value method for normalization operation;

[0042] Where x is the original feature parameter value; x′ is the normalized data; x min is the minimum value in a certain collected feature value data set; x′ is the maximum value in a certain feature data set.

[0043] In an optional solution: in step S5, a three-dimensional space point Pn is constructed based on the normalized absolute mean, root mean square, and variance, and the values ​​in the three dimensional directions of the point represent the absolute mean S abs , root mean square S rms , variance S δ In step S4, the normalized numerical points are collected and the n data points Pn are normalized by the formula It is divided into two categories, indicating that the pump is in normal operation and abnormal operation respectively, and the values ​​of the two data center points C1 and C2 are found through the evaluation function J. For the evaluation function J, when the target set S is input, the evaluation function J has a minimum value, and the point located in the evaluation function J satisfies the formula

[0044] Where J is the minimum average error cumulative sum of all mathematical objects; Pn is a point in the vector space Rd; ci is the cluster center point, ||p n -c i ||2 is the Euclidean distance from the nth point to the i-th cluster center.

[0045] In an optional solution: in step S6, the distance from Pn to the i cluster center points ci is recorded as di, the obtained i numerical points with a distance of di are compared, the smallest value is selected from them, and the point corresponding to the minimum value is recorded as point ci, which represents the operating status at a certain moment t.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. The present invention uses the K-means numerical algorithm to scientifically distinguish between normal operating conditions and abnormal operating conditions, without the need to set the threshold value of the abnormal pump vibration state based on experience.

[0048] 2. The present invention is based on a hidden Markov model over a period of time, and distinguishes pump unit system failures in the form of the cumulative product of probabilities over a period of time, rather than relying on the state at a certain moment, effectively avoiding the impact of random errors on the system, reducing misjudgments, and improving the system state recognition rate and accuracy.

[0049] 3. The model of the present invention is updatable and can make full use of the signal data accumulated in the actual production process. By conveniently updating the training data set, the model of the core part of the system can be updated. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a general structural diagram of the pump intelligent status monitoring system test platform corresponding to the present invention.

[0051] Figure 2 This is a signal diagnosis flow chart of the present invention.

[0052] Figure 3 This is a diagram showing the pump operation status classification effect achieved by the present invention based on the K-means cluster analysis algorithm.

[0053] Figure 4 This is the result diagram of the status identification of 10 pumps according to the present invention.

[0054] Figure 5 Schematic diagram of pump pressure transfer probability in healthy state based on historical data statistics.

[0055] Figure 6 Schematic diagram of pump pressure transfer probability in a decaying state obtained based on historical data statistics.

[0056] Figure 7 This is the probability diagram of the pump being in different pressure states based on historical statistics.

[0057] Figure 8 It is a state probability numerical diagram calculated based on the present invention. DETAILED DESCRIPTION

[0058] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0059] In one embodiment, Figures 1-8 As shown in Figure 2, a probability-driven pump unit system degradation identification method is proposed.

[0060] Includes offline and online stages;

[0061] The offline phase includes the following steps:

[0062] S1 pump vibration signal measurement:

[0063] When measuring pump vibration, place a vibration acceleration sensor at the pump head and retain the vibration acceleration signal measured in the Z direction. Place the pump unit under test on the test platform within the test bench to obtain vibration signals that exhibit both abnormal and normal operating conditions. These target characteristics can be verified using a pressure sensor, flow sensor, and laser velocimeter. Measure 200 sets of pump vibration acceleration signals Sn that exhibit these target conditions. Each set of signals is measured for 1 second at a sampling frequency of 10,000 Hz.

[0064] S2 pump signal filtering:

[0065] The 200 groups of original signals with a detection duration of 1s were put into the wavelet packet formula shown below for signal processing. The Bior3.7 wavelet was selected as the mother wave function, and the decomposition frequency band and layer number were j = 9, k = 12ψ * (·) is the conjugate function of ψ(·), according to the formula The original signal is processed to obtain 200 new sub-band signals S (j = 9, k = 12), which are re-recorded as signal S n (n=1,2,3…200), save the signal.

[0066] S3 pump signal feature extraction:

[0067] The 200 groups of sub-band discrete signals obtained by processing and containing 10,000 discrete data points are preferably converted into characteristic values ​​containing 200 groups of three characteristic parameters (a total of 600 numerical points) of absolute mean, root mean square, and variance according to formulas (18)-(20), and then substituted into Where m=10000.

[0068] S4 data normalization:

[0069] The three acquired eigenvalues ​​Sabs, Srms, and Sδ are calculated according to the formula The maximum value method is used for normalization operation, where: x is the original characteristic parameter value; x′ is the normalized data; x min is the minimum value in a certain collected feature value data set; x′ is the maximum value in a certain feature data set.

[0070] S5 evaluation function construction:

[0071] The normalized absolute mean, root mean square, and variance values ​​in a certain state are used to form a three-dimensional space point Pn in the three dimensions of X, Y, and Z. The 200 sets of data points Pn collected are calculated by the formula It is divided into two categories, indicating that the pump is in normal operation state and abnormal operation state respectively, and the formula Evaluation function J finds the values ​​of the two data center points C1 and C2. For evaluation function J, when the target set S is input, evaluation function J has a minimum value, and the points in evaluation function J satisfy

[0072] Where: J is the minimum average error cumulative sum of all mathematical objects; Pn is the point in the vector space Rd; ci is the cluster center point, ||p n -c i ||2 is the Euclidean distance from the nth point to the i-th cluster center. After calculation, the characteristic state cluster points of the normal operation state and the abnormal operation state are (0.82, 0.81, 0.71) and (0.23, 0.24, 0.11).

[0073] S6 status classification:

[0074] Through steps S1-S5, the cluster centers of the classification are found. The target's current operating status is determined based on the distance between the normalized numerical point and the cluster center. Let the distance from pn to the i cluster centers Ci be di. Compare the i numerical points with a distance of di and select the smallest value. The operating status represented by the Ci point corresponding to the minimum value is the operating status at a certain time t.

[0075] Through steps S1-S5, the offline phase completes the data set construction and status distinction of the normal operation state and abnormal operation state of the pump to be tested.

[0076] The 200 groups of data points, cluster centers, and state classification results calculated from steps S3-S5 are shown in the figure below. Figure 3 shown.

[0077] The online phase includes the following steps:

[0078] S7 online phase signal measurement:

[0079] A vibration sensor is placed on the pump unit to be tested. The signal collected at time t (t = 1s, 2s, 3s, ...) is recorded as Sn (n = 1s, 2, 3, ... 360). Continuously collect 360 segments of the original signal Sn at a frequency of 10,000 Hz over 360 seconds. Following steps S1-S4, the original signal is filtered, feature extracted, and normalized to obtain several groups of feature data points pn containing absolute mean, root mean square, and variance information. These data points are then compared with the previously trained cluster centers of normal operation (0.82, 0.81, 0.71) and abnormal operation (0.23, 0.24, 0.11). The value with the smallest Euclidean distance is selected as the state reflected at a specific time t.

[0080] For the 10 pumps tested, the test results obtained within 360 seconds according to the above method are as follows: Figure 4 shown

[0081] S8 state probability model construction:

[0082] Review the state probability data of the magnetic pump in the healthy and decaying states. For example, in the test process of this example, the initial state of the magnetic pump obtained by statistics is constant at high pressure, so its initial state probability matrix is ​​defined as π = π f =[1 0 0] 1×3 ; The probability of pressure transfer state caused by the health status of the pump in healthy state / degradation process is as follows: Figure 5 and Figure 6 As shown, for a healthy pump, the probability matrix of its corresponding pressure state changing to a specified pressure state is defined as For a pump in a decaying state, the probability of pressure change can be defined as

[0083] like Figure 7 As shown in the figure, the probability results of the pump operating normally / abnormally due to pressure changes are obtained based on historical data statistics. Therefore, the probability matrix of specific operating conditions caused by pressure changes is defined as Based on the above parameters, the hidden Markov model parameter λ of the healthy pump can be established h =(A, B, π) and the hidden Markov model parameter λ of the decay state pump f =(A f ,B f ,π), which is used for subsequent calculations.

[0084] S9 pump operating state probability calculation:

[0085] The following formula for the pump to appear in a certain state P t (O t |λ)=πAt-1 B;

[0086] Calculate the probability matrix of a certain operating state of the pump unit at time t, such as the pump in a healthy state at 1s, P1(O1|λ h )=πB; at 2s, P2(O2|λ h )=πAB;at 3s,P3(O3|λ h )=πA 2 B, and so on, at the 360th second, the probability matrix representing healthy operation and abnormal operation can be expressed as P 360 (O 360 |λ h )=πA 359 B;

[0087] Similarly, for the pump in decay state running for 1s, calculate the operating state probability matrix P1(O1|λ f )=πB f , at 2s, the probability is P2(O2|λ f )=πA f B f , at 3s, the probability is P3(O3|λ f )=πA f 2 B f , and so on, then at the 360th second, the probability matrix of healthy operation and abnormal operation can be expressed as: P 360 (O 360 |λ)=πA f 359 B f The corresponding calculation results are as follows: Figure 8 shown.

[0088] S10 pump health status judgment:

[0089] Depend on Figure 4 The pump operating status identified in the above equations assumes that the tested pump is a healthy pump and a decaying pump respectively; according to the formula Multiply the state probabilities identified by the pump unit at each corresponding moment within 360 seconds to obtain the probability value of the pump generating the determined state sequence within 360 seconds. t (O|λ h ) represents the state sequence probability result assuming that the pump unit to be tested is healthy, P t (O|λ f ) represents the state sequence probability result of the pump unit under test when it is decaying. If P t (O|λ) h >P t (O|λ) dIf the pump unit is in a healthy state, the pump unit is judged to be in a degraded state.

[0090] like Figure 1 As shown, Figure 1 The device includes a computer for calculating and processing data during the implementation of the present invention. A data acquisition card is used to obtain signals from corresponding signal sensors and transmit them to the computer. Among the sensors, a laser velocimeter can be used to obtain the pump's speed signal, a pressure sensor can be used to monitor the pressure at the pump's inlet and outlet, and a rotor flowmeter can be used to monitor the pump's flow signal. A vacuum pump is used to control the pressure within the system. These instruments can be used to assist in determining the operating status of the pump unit, and an acceleration sensor is used to monitor the pump's vibration signal.

[0091] When the pump is operating normally, it will usually show a certain amplitude value within a certain period. If the amplitude shows multiple abnormal amplitudes within a certain period of time and exceeds the threshold probability of a given operating state, then the pump can be determined to be operating abnormally.

[0092] like Figure 2 As shown, the method of the present invention specifically includes the following two parts, wherein steps S1-S7 are the offline database construction stage, which is used to collect and train the pump vibration signal, and steps S8-S16 are the online stage, which is used to diagnose the health status of the pump.

[0093] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A probability-driven pump unit system degradation identification method, characterized in that: Includes offline and online stages; The offline phase includes the following steps: S1: Conduct vibration signal acquisition test on the pump unit; S2: performing signal processing on a plurality of groups of original vibration signals collected within a unit time to obtain a plurality of groups of sub-band discrete signals, and saving them as signals to be processed; S3: extracting features from the collected signals of several groups of sub-band discrete signals containing several discrete data points to obtain three original feature parameter values; S4: The three original characteristic parameter values ​​collected are normalized using the maximum value method to obtain three normalized value points; S5: constructing three-dimensional space points based on the one-to-one correspondence between the three normalized numerical points and the state, and dividing the collected three-dimensional space points into two categories, representing the normal operation state and abnormal operation state of the pump, respectively. Then, the values ​​of the data center points of the two categories are found to obtain the cluster center points; S6: Based on the distance between the normalized numerical point and the cluster center, the target's current operating status is distinguished, and the normal operation state and abnormal operation state of the pump are automatically classified to complete the construction of the offline stage data set.

2. A probability-driven pump unit system degradation identification method according to claim 1, characterized in that: The online phase includes the following steps: S7: Install a vibration sensor on the pump to be tested, measure the vibration acceleration signal of the pump over a period of time, and perform signal processing, feature extraction, and normalization operations on the obtained raw signal in the same manner as steps S1-S4 to obtain several groups of numerical points. These groups of numerical points are compared with the cluster center points of the normal operation state and abnormal operation state of the pump trained in step S5; S8: Collect health and decay status probability data for the same type of pumps, and calculate the initial state probability P1, pressure change probability P2 caused by the health status of the pump, and normal and abnormal operation probabilities P3 caused by pressure state changes of the same type of pumps, and integrate the data into a hidden Markov model λ = (A, B, π); S9: Construct a set of hidden states and observable states, Define Q as the set of all possible hidden states, V as the set of all possible observed states, N as the number of possible hidden states, and M as the number of all possible observable states; S10: Construct hidden state and observable state sequence, A sequence of length T, I is the corresponding hidden state sequence, O is the corresponding observable sequence, any hidden state i t ∈Q, any observation state o t ∈V; S11: Construct hidden state probability matrix: The hidden state at time t is i t =q i , the hidden state at time t+1 is i t+1 =q j , the state transition probability a from time t to time t+1 ij It can be expressed as: ij =P(i t+1 =q j |i t =q i ); a ij The hidden state Markov state transition matrix A can be formed: A=[a ij ] N×N ; S12: Construct observable state probability matrix: The hidden state at time t is i t =q j , the corresponding observation state is o t =v k , the state v at this moment k In the hidden state q j The probability of generating is b j (k), satisfies: b i (k)=P(o t =v k |i t =q j ); b j (k) can form the probability matrix B of the observed state generation: B = [b j (k)] N×M S13: Construct the initial state probability matrix: The hidden state probability distribution π at the initial time t=1 is defined as: π=[π(i)] N , where π(i)=P(i1=q i ); S14: Calculate the probability of unobservable state: For a given observation sequence O, the probability of an unobservable feature it appearing at time t is calculated as: P t (i t |λ)=πA t-1 ; S15: Calculate the probability of observable state: Calculate the probability of a certain observable state Ot occurring at time t as P t (O t |λ)=πA t-1 B; S16: Calculate the state sequence probability: At time t, for a given observation sequence O, the probability of occurrence of the state sequence is S17: Perform state probability comparison, record the corresponding vibration signals of the pump within a period of time t, calculate and identify the operating state of the pump at each moment in the time period in step S7, and obtain the state sequence of the pump unit within the period of time t based on the identified operating state; Assuming that the pump is in a healthy state and a decaying state, calculate the probability Pt(O|λh) of the state sequence when the pump unit is in a healthy state and the probability Pt(O|λf) of the state sequence when the pump unit is in a decaying state according to steps S10-S16, and compare the value of Pt(O|λh) with Pt(O|λf). If P t (O|λ h )>P t (O|λ d ) the pump unit under test is judged to be in a healthy state; otherwise, the pump unit is judged to be in a decayed state.

3. The probability-driven pump unit system degradation identification method according to claim 1 is characterized in that: In step S2, suppose that the original vibration signals collected in the unit time t are s(t), and the wavelet packet formula is used for signal processing, and the Bior3.7 wavelet is selected as the mother wave function, ψ * (·) is the conjugate function of ψ(·), according to the formula Process the original signal to obtain a new signal S(j,k); Where j and k represent the scale parameter and translation parameter respectively; ψ * (·) represents the size of the mother wave ψ(·) and the transformation mother wave function; j and k represent the scale parameter and translation parameter respectively; the formula is used. The 12th group of sub-band signals S (j=9, k=12) of the 9th layer are saved as filtered signals to be processed.

4. A probability-driven pump unit system degradation identification method according to claim 1, characterized in that: In step S3, the collected signals Sn of several groups of sub-band discrete signals and containing m discrete data points are processed. The three formulas are used to obtain the absolute mean S abs , root mean square S rms , variance S δ These three characteristic parameters are used as original characteristic parameter values.

5. The probability-driven pump unit system degradation identification method according to claim 1 is characterized in that: In step S4, the three original characteristic parameter values ​​collected are The formula uses the maximum value method for normalization; Where x is the original feature parameter value; x′ is the normalized data; x min is the minimum value in a certain collected feature value data set; x′ is the maximum value in a certain feature data set.

6. A probability-driven pump unit system degradation identification method according to claim 1, characterized in that: In step S5, a three-dimensional space point Pn is constructed based on the normalized absolute mean, root mean square, and variance. The values ​​in the three dimensional directions of the point represent the absolute mean S abs , root mean square S rms , variance S δ In step S4, the normalized numerical points are collected and the n data points Pn are normalized by the formula It is divided into two categories, indicating that the pump is in normal operation and abnormal operation respectively, and the values ​​of the two data center points C1 and C2 are found through the evaluation function J. For the evaluation function J, when the target set S is input, the evaluation function J has a minimum value, and the point located in the evaluation function J satisfies the formula Where J is the minimum average error cumulative sum of all mathematical objects; Pn is a point in the vector space Rd; ci is the cluster center point, ||p n -c i ||2 is the Euclidean distance from the nth point to the i-th cluster center.

7. A probability-driven pump unit system degradation identification method according to claim 6, characterized in that: In step S6, the distance from Pn to the i cluster center points ci is recorded as di, and the obtained i numerical points with a distance of di are compared, and the smallest value is selected from them. The point corresponding to the minimum value is recorded as point ci, and the operating state represented by it is the operating state at a certain time t.

Images