Fault diagnosis method for tobacco shred vibrating conveyor

Abstract

The invention discloses a fault diagnosis method for a tobacco shred vibration conveyor, which comprises the following steps: arranging a vibration sensor at a preset position of the vibration conveyor to collect vibration signals of the vibration conveyor in different working states; decomposing the collected vibration signal based on an improved ensemble empirical mode decomposition (EEMD) algorithm; on the basis of kurtosis and correlation coefficient criteria, IMF components obtained through decomposition are screened, then signals are reconstructed, and feature vectors are constructed; and dividing the constructed feature vector into a training sample and a test sample, inputting the training sample as an observation sequence into a hidden Markov model (HMM) for training to obtain an improved EEMD-HMM model, and inputting the test sample into the improved EEMD-HMM model for state recognition to complete fault diagnosis of the vibrating conveyor. According to the invention, the accuracy of fault type identification can be improved, the production cost is reduced, and the maintenance time is shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis of general tobacco conveying equipment, and more specifically, to a fault diagnosis method for a tobacco vibrating conveyor. Background Art

[0002] The conveying equipment for tobacco is mainly belt conveyors and vibration conveyors. The conveying capacity of belt conveyors is very stable, but due to its vibrating working state, the conveying capacity of vibration conveyors is often affected by abnormal working conditions of their parts. Wear of vibration conveyor motor pulleys, wear of shock-absorbing sleeves, and fatigue failure of shock-absorbing springs will affect the dynamic balance, reduce the conveying capacity of the vibration conveyor, and easily cause blockage. Once blocked, tobacco will be exposed to the air for a long time, seriously affecting the filling capacity and sensory quality of tobacco. Therefore, it is of great significance to make timely warnings for abnormal working conditions of vibration conveyors and accurately locate fault locations to reduce production costs and reduce maintenance time. Summary of the invention

[0003] The present invention provides a fault diagnosis method for a tobacco vibrating conveyor, which solves the problem that the existing tobacco vibrating conveyor cannot obtain early warning and diagnosis in time when a fault occurs, which easily causes material blockage. The method can improve the accuracy of fault type identification, reduce production costs, and shorten maintenance time.

[0004] To achieve the above objectives, the present invention provides the following technical solutions:

[0005] A method for diagnosing a fault of a tobacco vibrating conveyor, comprising:

[0006] A vibration sensor is arranged at a preset position of the vibration conveyor to collect vibration signals of the vibration conveyor under different working conditions;

[0007] Decomposing the collected vibration signal based on an improved ensemble empirical mode decomposition algorithm EEMD;

[0008] The IMF components obtained by decomposition are screened based on the kurtosis and correlation coefficient criteria, and then the signal is reconstructed and the feature vector is constructed;

[0009] The constructed feature vectors are divided into training samples and test samples. The training samples are input into the hidden Markov model HMM as observation sequences for training to obtain the improved EEMD-HMM model. The test samples are input into the improved EEMD-HMM model for state recognition to complete the fault diagnosis of the vibration conveyor.

[0010] Preferably, the collected vibration signals of the vibration conveyor under different working conditions include: vibration signals under three working conditions: wear of the vibration conveyor motor pulley, wear of the shock absorbing sleeve and fatigue failure of the shock absorbing spring.

[0011] Preferably, the decomposing the collected vibration signal based on an improved ensemble empirical mode decomposition algorithm EEMD comprises:

[0012] The bacterial foraging algorithm (BFA) is used to optimize the parameters of the ensemble empirical mode decomposition algorithm (EEMD). The standard deviation of the EEMD decomposition result is used as the objective function. After the trending operation, replication operation and migration operation, the fitness value is calculated, compared and updated to find the best combination of the two most important parameters affecting the EEMD decomposition effect, the white noise amplitude coefficient Nstd and the overall average number NE.

[0013] The vibration signal is decomposed by optimized EEMD, and the vibration signal of each state will be decomposed to obtain a series of IMF components and a residual component.

[0014] Preferably, the method of using the bacterial foraging algorithm BFA to perform parameter optimization on the ensemble empirical mode decomposition algorithm EEMD includes:

[0015] For the population size S and the number of trend operations in BFA , Copy operation count , Number of migration operations , migration probability , Maximum number of steps for one-way movement in trend operation Initialize the parameters;

[0016] Import the vibration signal data, and then randomly generate S sets of (Nstd, NE) coordinates as the initial positions of the bacteria;

[0017] Substitute each set of coordinates into the fitness function for calculation and obtain the standard deviation.

[0018] Preferably, the screening of the IMF components obtained by decomposition based on the kurtosis and correlation coefficient criteria, and then reconstructing the signal and constructing the feature vector comprises:

[0019] Assume that any IMF component obtained by decomposing the signal is a discrete sequence , then this discrete sequence The kurtosis formula is:

[0020] ,in, For discrete sequences The mean of For discrete sequences The standard deviation of n is a discrete sequence The number of signals included;

[0021] Assume that the original signal collected is a discrete sequence , any IMF component obtained by decomposing the original signal is , then the correlation coefficient between the two is:

[0022] , where u represents a discrete sequence and The number of signals included, Represents a discrete sequence The average value of Represents a discrete sequence The average value of

[0023] Calculate the kurtosis and correlation coefficient of each IMF component, select the first five components with large kurtosis and correlation coefficient, and define the selected components as , where r is the serial number of the screened component, b is the serial number of the element contained in the screened component, and the signal is reconstructed;

[0024] The energy of each component in the reconstructed signal is defined as , find the energy corresponding to all the screened components, and then sum them up, that is:

[0025] , where R is the total number of components screened by the kurtosis and correlation coefficient criteria;

[0026] Find the ratio of the energy of each component to the total energy, that is, the energy proportion, and construct the energy characteristic sequence X, that is: ;

[0027] The Lloyds algorithm is used to scalar quantize the energy feature sequence X, and finally the feature vector is obtained.

[0028] Preferably, dividing the constructed feature vector into training samples and test samples comprises:

[0029] The constructed feature vectors are divided into training samples and test samples according to a 5:5 ratio.

[0030] Preferably, the test sample is input into the improved EEMD-HMM model for state recognition to complete the fault diagnosis of the vibrating conveyor, including:

[0031] The feature vector of the vibration conveyor is input into the trained improved EEMD-HMM model. At this time, each model will calculate a log-likelihood probability based on the feature vector. The state corresponding to the model with the largest calculated log-likelihood probability is the current fault type of the vibration conveyor.

[0032] Preferably, the fault types include: wear of the motor pulley, wear of the shock-absorbing sleeve, and fatigue failure of the shock-absorbing spring.

[0033] The present invention provides a fault diagnosis method and system for a tobacco vibrating conveyor, proposes a model for optimizing EEMD using BFA (bacterial foraging algorithm), decomposes and compares vibration signals using the models before and after optimization, verifies the superiority of the optimization model in terms of the number of components, reconstruction error and decomposition efficiency, and finally, uses a model combining BFA-EEMD and HMM for training and recognition comparison to accurately identify the fault type, thereby solving the problem that the existing tobacco vibrating conveyor cannot obtain early warning and diagnosis in time when a fault occurs, which easily causes material blockage, can improve the accuracy of fault type identification, reduce production costs, and reduce maintenance time. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the specific embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below.

[0035] Figure 1 The present invention provides a flow chart of a method for diagnosing faults of a tobacco vibrating conveyor.

[0036] Figure 2 The following are vibration signal waveforms of three different fault types of a vibrating conveyor collected in an embodiment of the present invention.

[0037] Figure 3 Parameter optimization diagram of BFA optimizing EEMD in the embodiment of the present invention

[0038] Figure 4 This is a comparison diagram of the decomposition effects of optimized EEMD and non-optimized EEMD in the embodiment of the present invention.

[0039] Figure 5 This is a comparison diagram of optimized EEMD and non-optimized EEMD decomposition iterations in an embodiment of the present invention.

[0040] Figure 6 This is a flow chart of HMM model training in an embodiment of the present invention.

[0041] Figure 7 This is a comparison diagram for identifying wear of a motor pulley of a vibrating conveyor in an embodiment of the present invention.

[0042] Figure 8 This is a comparison diagram for identifying wear of the shock-absorbing sleeve of a vibrating conveyor in an embodiment of the present invention.

[0043] Fig. 9 This is a comparison diagram for fatigue failure identification of a vibration conveyor damping spring in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to enable persons skilled in the art to better understand the solutions of the embodiments of the present invention, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and implementation modes.

[0045] In view of the problem that the current tobacco vibrating conveyor cannot obtain timely warning and diagnosis when a fault occurs, which easily causes material blockage, the present invention provides a fault diagnosis method and system for a tobacco vibrating conveyor, which solves the problem that the existing tobacco vibrating conveyor cannot obtain timely warning and diagnosis when a fault occurs, which easily causes material blockage. The method can improve the accuracy of fault type identification, reduce production costs, and shorten maintenance time.

[0046] like Figure 1 As shown, a fault diagnosis method for a tobacco vibrating conveyor comprises:

[0047] S1: A vibration sensor is set at a preset position of the vibration conveyor to collect vibration signals of the vibration conveyor under different working conditions.

[0048] S2: Decomposing the collected vibration signal based on an improved ensemble empirical mode decomposition algorithm EEMD.

[0049] S3: The decomposed IMF components are screened based on the kurtosis and correlation coefficient criteria, and then the signal is reconstructed and the feature vector is constructed.

[0050] S4: The constructed feature vector is divided into training samples and test samples. The training samples are input into the hidden Markov model HMM as observation sequences for training to obtain an improved EEMD-HMM model. The test samples are input into the improved EEMD-HMM model for state recognition to complete the fault diagnosis of the vibration conveyor.

[0051] Specifically, in step S1, the collected vibration signals are vibration signals under three working conditions: wear of the vibration conveyor motor pulley, wear of the shock absorbing sleeve, and fatigue failure of the shock absorbing spring. Figure 2 shown.

[0052] In step S2, BFA (bacterial foraging algorithm) is used to optimize the parameters of EEMD (ensemble empirical mode decomposition), and the standard deviation of the EEMD decomposition result is used as the objective function. After undergoing trend operation, replication operation and migration operation, the fitness value is calculated, compared and updated to find the optimal combination of the two most important parameters affecting the EEMD decomposition effect, the white noise amplitude coefficient (Nstd) and the overall average number (NE), and the improved EEMD is obtained.

[0053] For the population size S and the number of trend operations in BFA , Copy operation count , Number of migration operations , migration probability , Maximum number of steps for one-way movement in trend operation Initialize the parameters.

[0054] The vibration signal data is imported, and then S groups of (Nstd, NE) coordinates are randomly generated as the initial positions of the bacteria.

[0055] Each set of coordinates is brought into the fitness function for calculation to obtain the standard deviation (i.e., fitness value), and the fitness value is continuously calculated, compared, and updated.

[0056] In the algorithm , and After reaching the initial set number of times, the minimum fitness value will be found. At this time, the optimal parameter combination can be obtained according to the coordinate position of the bacteria, such as Figure 3 As shown, substituting the optimal parameter combination into EEMD can obtain the improved EEMD.

[0057] In step S3, the process of improving EEMD to decompose the vibration signal is as follows:

[0058] ① Definition is the original vibration signal, and a white noise sequence is added to the original vibration signal , get a new signal ,Right now:

[0059]

[0060] Where: represents the white noise added for the ith time, Represents the new signal obtained after adding white noise for the i-th time.

[0061] ② For new signals By performing EMD decomposition, we can obtain a series of IMF components. and the residual component ,Right now:

[0062]

[0063] Where: is the jth IMF component obtained by adding white noise for the i-th time, is the residual component.

[0064] ③. Repeat steps ① and ② N times. Then, in order to offset the influence of white noise on the original vibration signal, perform mean processing on each group of IMF components decomposed in the above steps. Finally, the IMF components after improved EEMD decomposition can be obtained. and residual ,Right now:

[0065]

[0066]

[0067] Where: represents the jth component of the original vibration signal obtained by improved EEMD decomposition, represents the final residual component.

[0068] ④ Final original vibration signal After improved EEMD decomposition, M IMF components and one residual component can be obtained, namely:

[0069]

[0070] In order to verify the superiority of the improved EEMD decomposition effect, the same set of vibration signals are decomposed using the improved EEMD and the unimproved EEMD. The decomposition comparison results are shown in Figure 2. Figure 4 As shown in the figure, after optimized EEMD decomposition, the vibration signal obtains eight-order components IMF1~IMF8 and a residual component IMF9, and the time taken for signal decomposition is 1.3808s, while after unoptimized EEMD decomposition, the tenth-order components IMF1~IMF10 and a residual component IMF11 are obtained, and the time taken for signal decomposition is 3.5494s. The reduction in the number of decomposed components and the reduction in the time taken for decomposition prove that the improved EEMD decomposition is more efficient and can more adaptively decompose the added signal.

[0071] In order to further illustrate that the improved EEMD can more adaptively decompose the added signal and has a higher decomposition efficiency, the improved EEMD and the computational cost required for decomposing the signal are shown in the form of a box plot, as shown in Figure 5 As shown. The horizontal axis in the figure represents the number of each component, and the vertical axis represents the number of iterations required to decompose each IMF. After calculation, it can be seen that the number of iterations required to decompose the early wear vibration signal of the tool by the unoptimized EEMD to obtain the eleventh-order component is 10357 times, while the number of iterations required to decompose the ninth-order component by the optimized EEMD is 3728 times. The calculation cost is about 36% of the calculation cost of the unoptimized EEMD. This proves that the optimized EEMD has better adaptability and higher decomposition efficiency.

[0072] In step S4, it is assumed that any IMF component obtained by decomposing the signal is a discrete sequence , then this discrete sequence The kurtosis formula is:

[0073]

[0074] Where: For discrete sequences The mean of For discrete sequences The standard deviation of n is a discrete sequence The number of signals included.

[0075] In the vibration signal, the impact component is proportional to the kurtosis value. In order to ensure the richness of the fault information in the reconstructed signal, the component with a larger kurtosis value is screened out. Here, the vibration signal collected when the motor pulley of the vibration conveyor is worn is taken as an example. The components of the vibration signal are decomposed by the improved EEMD. The kurtosis values ​​of each component are calculated as shown in Table 1.

[0076]

[0077] Assume that the original vibration signal collected is a discrete sequence , any IMF component obtained by decomposing the original vibration signal is , then the correlation coefficient between the two is:

[0078]

[0079] Where: u represents a discrete sequence and The number of signals included, Represents a discrete sequence The average value of Represents a discrete sequence The average value of .

[0080] The strength of the correlation between variables is directly proportional to the size of the correlation coefficient. In order to retain the fault information to the greatest extent when reconstructing the signal, the components with larger correlation coefficients are selected. Here, the vibration signal collected when the motor pulley of the vibration conveyor is worn is taken as an example. The components of the vibration signal are decomposed by the improved EEMD. The correlation coefficients of each component and the original signal are calculated as shown in Table 2.

[0081]

[0082] Since the components are selected based on the criteria of large kurtosis and correlation coefficient, after comprehensive consideration of the data in Tables 1 and 2, the first five-order components IMF1~IMF5 are selected for signal reconstruction.

[0083] The filtered components are defined as , where r is the serial number of the screened component, b is the serial number of the element contained in the screened component, and the energy of each component is defined as , then the feature vector is constructed as follows:

[0084] Find the energy corresponding to all the filtered components and then sum them up, that is:

[0085]

[0086] Where: R is the total number of components screened by the kurtosis and correlation coefficient criteria.

[0087] Find the ratio of the energy of each component to the total energy, that is, the energy proportion, and construct the energy characteristic sequence X, that is:

[0088]

[0089] The reason why the energy of a single component and the total energy are used to calculate the ratio is that this method can avoid the interleaving distortion between different components and is conducive to extracting features from the signal.

[0090] The Lloyds algorithm in the source coding technology in the communication field is used to scalar quantize the obtained energy feature sequence X, and finally the feature vector is obtained.

[0091] The collected vibration signals under three working conditions, namely, motor pulley wear, shock-absorbing sleeve wear, and shock-absorbing spring fatigue failure of the vibration conveyor, were optimized for EEMD decomposition, and then the required components were screened out according to the kurtosis criterion and the correlation coefficient criterion. Finally, according to the above steps of constructing feature vectors, the feature vectors of the vibration conveyor in different fault states were constructed respectively. Due to the large amount of constructed feature vector data, only five groups of feature vectors of the vibration conveyor in different fault states are attached here, as shown in Table 3.

[0092]

[0093] 40 groups of samples of vibration conveyor motor pulley wear, shock absorber sleeve wear and shock absorber spring fatigue failure were collected respectively, for a total of 120 groups. These samples were decomposed according to the method mentioned above, that is, after optimizing EEMD, the components were screened for signal reconstruction and feature vectors were constructed. The first 20 groups of feature vectors of the three fault states were taken respectively, for a total of 60 groups, as training samples for HMM training. During model training, the maximum number of iterations L was set to 50, and the convergence error was 0. , the HMM training process is as follows Figure 6 shown.

[0094] After the HMM model is trained, the next step is to verify the model recognition effect. The recognition process is to input the feature vector of the vibration conveyor into the trained model. At this time, each model will calculate a log-likelihood probability based on the feature vector. The state corresponding to the model with the largest calculated log-likelihood probability is the current fault type of the vibration conveyor.

[0095] The last 20 groups of feature vectors of the three types of vibration conveyor faults constructed by the improved EEMD decomposition above, a total of 60 groups, are input into the model trained by the feature vector constructed by the improved EEMD decomposition for identification. At the same time, in order to verify that the decomposition effect of the improved EEMD is better, 20 groups of feature vectors of the signals of the three types of vibration conveyor faults are constructed by the unoptimized EEMD decomposition, a total of 60 groups, and input into the model trained by the feature vector constructed by the unimproved EEMD decomposition for diagnosis. The recognition results of the three types of vibration conveyor faults are compared as follows: Figure 7 , Figure 8 as well as Fig. 9 shown.

[0096] from Figure 7 , Figure 8 as well as Fig. 9 It can be seen that the feature vector constructed after the improved EEMD decomposition has a higher accuracy in vibration conveyor fault identification than the feature vector constructed after the unimproved EEMD decomposition. The specific accuracy is shown in Table 4.

[0097]

[0098] From the data in the table, it can be concluded that the accuracy of the feature vector constructed after the improved EEMD decomposition in the vibration conveyor fault identification is higher than the accuracy of the feature vector constructed after the unimproved EEMD decomposition in the vibration conveyor fault identification, which once again verifies the superiority of the improved EEMD decomposition.

[0099] It can be seen that the present invention provides a fault diagnosis method for a tobacco vibrating conveyor, proposes a model that uses BFA (bacterial foraging algorithm) to optimize EEMD, uses the models before and after optimization to decompose and compare the vibration signal, and verifies the superiority of the optimization model in terms of the number of components, reconstruction error and decomposition efficiency. Finally, a model combining BFA-EEMD and HMM is used for training and recognition comparison to accurately identify the fault type, solve the problem that the existing tobacco vibrating conveyor cannot be warned and diagnosed in time when a fault occurs, which easily causes blockage, can improve the accuracy of fault type identification, reduce production costs, and reduce maintenance time.

[0100] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the drawings. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the description and drawings, should be within the protection scope of the present invention.

Claims

1. A method for diagnosing faults of a tobacco vibrating conveyor, characterized in that: include: A vibration sensor is arranged at a preset position of the vibration conveyor to collect vibration signals of the vibration conveyor under different working conditions; Decomposing the collected vibration signal based on an improved ensemble empirical mode decomposition algorithm EEMD; The IMF components obtained by decomposition are screened based on the kurtosis and correlation coefficient criteria, and then the signal is reconstructed and the feature vector is constructed; The constructed feature vectors are divided into training samples and test samples. The training samples are input into the hidden Markov model HMM as observation sequences for training to obtain the improved EEMD-HMM model. The test samples are input into the improved EEMD-HMM model for state recognition to complete the fault diagnosis of the vibration conveyor.

2. The fault diagnosis method for a tobacco vibrating conveyor according to claim 1, characterized in that: The vibration signals of the vibration conveyor under different working conditions are collected, including vibration signals under three working conditions: wear of the vibration conveyor motor pulley, wear of the shock absorbing sleeve and fatigue failure of the shock absorbing spring.

3. The fault diagnosis method for a tobacco vibrating conveyor according to claim 2, characterized in that: Decomposing the collected vibration signal based on an improved ensemble empirical mode decomposition algorithm EEMD includes: The bacterial foraging algorithm (BFA) is used to optimize the parameters of the ensemble empirical mode decomposition algorithm (EEMD). The standard deviation of the EEMD decomposition result is used as the objective function. After the trending operation, replication operation and migration operation, the fitness value is calculated, compared and updated to find the best combination of the two most important parameters affecting the EEMD decomposition effect, the white noise amplitude coefficient Nstd and the overall average number NE. The vibration signal is decomposed by optimized EEMD, and the vibration signal of each state will be decomposed to obtain a series of IMF components and a residual component.

4. The fault diagnosis method for a tobacco vibrating conveyor according to claim 3, characterized in that: The method of using the bacterial foraging algorithm BFA to optimize the parameters of the ensemble empirical mode decomposition algorithm EEMD includes: For the population size S and the number of trend operations in BFA , Copy operation count , Number of migration operations , migration probability , Maximum number of steps for one-way movement in trend operation Initialize the parameters; Import the vibration signal data, and then randomly generate S sets of (Nstd, NE) coordinates as the initial positions of the bacteria; Substitute each set of coordinates into the fitness function for calculation and obtain the standard deviation.

5. The fault diagnosis method for a tobacco vibrating conveyor according to claim 4, characterized in that: The method of screening the IMF components obtained by decomposition based on the kurtosis and correlation coefficient criteria, and then reconstructing the signal and constructing the feature vector includes: Assume that any IMF component obtained by decomposing the signal is a discrete sequence , then this discrete sequence The kurtosis formula is: ,in, For discrete sequences The mean of For discrete sequences The standard deviation of n is a discrete sequence The number of signals included; Assume that the original signal collected is a discrete sequence , any IMF component obtained by decomposing the original signal is , then the correlation coefficient between the two is: , where u represents a discrete sequence and The number of signals included, Represents a discrete sequence The average value of Represents a discrete sequence The average value of Calculate the kurtosis and correlation coefficient of each IMF component, select the first five components with large kurtosis and correlation coefficient, and define the selected components as , where r is the serial number of the screened component, b is the serial number of the element contained in the screened component, and the signal is reconstructed; The energy of each component in the reconstructed signal is defined as , find the energy corresponding to all the screened components, and then sum them up, that is: , where R is the total number of components screened by the kurtosis and correlation coefficient criteria; Find the ratio of the energy of each component to the total energy, that is, the energy proportion, and construct the energy characteristic sequence X, that is: ; The Lloyds algorithm is used to scalar quantize the energy feature sequence X, and finally the feature vector is obtained.

6. The fault diagnosis method for a tobacco vibrating conveyor according to claim 5, characterized in that: The constructed feature vector is divided into training samples and test samples, including: The constructed feature vectors are divided into training samples and test samples according to a 5:5 ratio.

7. The fault diagnosis method for a tobacco vibrating conveyor according to claim 6, characterized in that: The test samples are input into the improved EEMD-HMM model for state recognition to complete the fault diagnosis of the vibration conveyor, including: The feature vector of the vibration conveyor is input into the trained improved EEMD-HMM model. At this time, each model will calculate a log-likelihood probability based on the feature vector. The state corresponding to the model with the largest calculated log-likelihood probability is the current fault type of the vibration conveyor.

8. The fault diagnosis method for a tobacco vibrating conveyor according to claim 7, characterized in that: The fault types include: motor pulley wear, shock-absorbing sleeve wear, and vibration-absorbing spring fatigue failure.

Images