Oiling machine diagnosis method based on combination of hidden Markov model and Viterbi algorithm

Abstract

The invention discloses an oiling machine diagnosis method based on the combination of a hidden Markov model and a Viterbi algorithm, and the method comprises the steps: building a corresponding state transition matrix and an observation matrix, building the hidden Markov model according to the operation condition of an oiling machine device, training the hidden Markov model according to historical data, and carrying out the diagnosis of the oiling machine through the Viterbi algorithm. The observation values generated in different states are distributed to a specific state, the transition probability between the states is established, and the most probable state sequence is found by using the Viterbi algorithm. According to the invention, real-time monitoring, prediction and maintenance are carried out on the whole life cycle of the oiling machine device, and harm caused by fault judgment lag is avoided.

Description

Technical Field

[0001] The invention relates to a diagnosis technology of a cold-rolled steel strip oiling machine device, and more specifically to an oiling machine diagnosis method based on a hidden Markov model combined with a Viterbi algorithm. Background Art

[0002] The cold-rolled strip oiler is a device used to produce steel coated with a protective oil film. Its working principle is to feed the untreated or otherwise pre-treated strip steel continuously, and spray and coat it at different nozzles on the upper and lower blade beams through the atomization of a high-pressure generator.

[0003] However, some failures may occur during the production process of the cold rolled strip oiling machine, such as the following:

[0004] Nozzle clogging: Due to the presence of particles in the coating material or the narrowing of the nozzle due to deposition, the flow rate decreases or the spraying stops.

[0005] Uneven oil coating thickness: Due to incorrect nozzle spray angle or unstable machine running speed, the oil coating thickness in some areas is too thick or too thin.

[0006] Transmission system failure: The transmission system mainly includes components such as reducers and chains. If these components are damaged or not replaced in time, it will lead to abnormal operation and reduced product quality.

[0007] Abnormal coating oil level: When the coating oil level is too low, it will affect the transmission of the equipment's autonomous monitoring system operation signal, causing production stagnation.

[0008] If the above faults are not discovered and handled in time, they may cause operational problems in the cold-rolled strip oiling machine, affecting product quality and production efficiency. However, at present, fault diagnosis relies on manual inspection of equipment, which is time-consuming, and some threatening fault points may be missed if the inspection is not standardized or the level is insufficient. Once a problem occurs, it is more likely to cause a production accident, and the fault diagnosis is delayed. Summary of the invention

[0009] In view of the defects existing in the prior art, the purpose of the present invention is to provide an oiler diagnosis method based on a hidden Markov model combined with a Viterbi algorithm, so as to realize real-time monitoring and predictive maintenance of the oiler device throughout its life cycle and avoid the harm caused by delayed fault judgment.

[0010] To achieve the above object, the present invention adopts the following technical solution:

[0011] A method for oiler diagnosis based on hidden Markov model combined with Viterbi algorithm:

[0012] By establishing the corresponding state transfer matrix and observation matrix, a hidden Markov model is established for the operation of the oiler device. The hidden Markov model is trained according to historical data. The observation values ​​generated under different states are assigned to a specific state, and the transition probability between each state is established. The Viterbi algorithm is used to find the most likely state sequence.

[0013] Preferably, the oiler diagnostic method specifically comprises the following steps:

[0014] S1, collecting status data of the oiling machine device;

[0015] S2, collecting variable data of the oil coating device;

[0016] S3, feature extraction;

[0017] S4, training the hidden Markov model;

[0018] S5, predicting the state of the oiler device in real time;

[0019] S6, troubleshooting the oiler device;

[0020] S7. Continuous optimization.

[0021] Preferably, the status data in step S1 includes normal operation, flow meter failure, inverter failure, metering pump failure, and low tank level.

[0022] Preferably, the variable data in step S2 include inverter output current, inverter fault information, actual speed of metering pump, temperature of metering pump, mailbox liquid level, flow meter flow, unit strip speed, and oiling set value.

[0023] Preferably, the feature extraction in step S3 specifically includes:

[0024] The hidden Markov model is used to extract features from the data collected in step S1 and step S2, and corresponding features are designed according to the operating status of the oiler device.

[0025] Preferably, the step S4 of training the hidden Markov model specifically includes:

[0026] The hidden Markov model is used to train and extract features, and the parameters of the hidden Markov model are estimated.

[0027] Preferably, the parameters of the hidden Markov model include an initial state probability vector PI, a transition probability matrix, and an observation probability matrix.

[0028] Preferably, the step S5 of predicting the state of the oiler device in real time specifically includes:

[0029] The most likely state sequence is inferred according to a given probability matrix by the Viterbi algorithm, and when an abnormality is detected, the corresponding state sequence is output.

[0030] Preferably, the step S6 of troubleshooting the oiler device specifically includes:

[0031] Confirm the fault location, analyze the cause of the fault and record it. Confirm the fault location found, analyze the cause, find ways to repair or adjust it, and record it in the database.

[0032] Preferably, the step S7 continuous optimization specifically includes:

[0033] Improve the parameters of the hidden Markov model, update and improve the training data of the hidden Markov model according to the data feedback such as the cause of the fault and the loss of components, and continuously adjust the parameters of the hidden Markov model.

[0034] The oiler diagnosis method based on hidden Markov model combined with Viterbi algorithm provided by the present invention has the following beneficial effects:

[0035] 1) High accuracy: The hidden Markov model can form a more accurate and comprehensive description of the state of the oiler device, so as to better diagnose faults. At the same time, the Viterbi algorithm can find the most likely state sequence, thereby improving the diagnostic accuracy;

[0036] 2) Strong flexibility: The hidden Markov model can be flexibly adjusted and optimized as needed to adapt to the working conditions of the oiling machine in actual production. For example, as the level of detection becomes more detailed, the coverage of the system can be further expanded;

[0037] 3) High efficiency: Since the oiler diagnostic method of the present invention uses a dynamic programming algorithm to calculate the state sequence, the operation speed is relatively fast;

[0038] 4) Reliable solution: The oiler diagnostic method of the present invention combines the theories of statistics and machine learning, can adapt to new environments by effectively utilizing historical data, and has high reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic flow chart of the oiler diagnostic method of the present invention. DETAILED DESCRIPTION

[0040] In order to better understand the above technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0041] Hidden Markov Model (HMM) is a common probabilistic statistical model, which is widely used in signal processing, natural language processing, speech recognition and other fields. HMM can be regarded as a two-sequence random process consisting of two sequences: observation variables and hidden state variables. The observation variables represent visible data, while the hidden state variables are not directly observed.

[0042] In HMM, hidden state variables are changed according to the transition probability distribution. Based on these changes, the hidden state variables will output a series of observations, where each observation has a certain probability associated with each hidden state variable.

[0043] HMM is often used in industrial systems in areas such as fault diagnosis, prediction, and condition monitoring. For example, in wind turbine equipment detection, HMM can be used to build a suitable model to process the generated data.

[0044] The Viterbi algorithm is a dynamic programming algorithm used to find the most likely state sequence in a hidden Markov model given a sequence of observed data.

[0045] In the Hidden Markov Model, the state of the system cannot be observed directly, but by observing a series of observable variables of the system, it can be inferred that the system is in different states at different times. The Viterbi algorithm mainly calculates the maximum probability value of each possible state at each moment, and gradually derives the state sequence with the maximum final probability value.

[0046] Specifically, the Viterbi algorithm includes the following three basic steps:

[0047] 1. Initialization: Calculate the probabilities of all possible states at the first moment.

[0048] 2. Recursion: For each time step t, calculate the maximum probability of each possible state at the current moment, and record the state of the previous moment that produced this probability until the last moment of the observed data sequence.

[0049] 3. Backtracking: Based on the state at the last moment and the state at the previous moment that generated it, backtrack forward to obtain the final estimated state sequence.

[0050] Combination Figure 1 As shown, the present invention provides a method for diagnosing an oiler based on a hidden Markov model combined with a Viterbi algorithm:

[0051] First, by establishing the corresponding state transfer matrix and observation matrix, the operation of the oiler device is modeled as a hidden Markov model. Then, the hidden Markov model is trained based on historical data, and the observations generated under different states are assigned to a specific state, and the transition probability between each state is established. In this way, the HMM can be used to predict the next state and observation results in an unknown state.

[0052] Secondly, the Viterbi algorithm is used to find the most likely state sequence. Given an observation sequence, the Viterbi algorithm is an effective dynamic programming algorithm that can calculate the most likely state sequence corresponding to the observation sequence. Therefore, this algorithm is very practical in the fault diagnosis of the oiling machine device.

[0053] The oiler diagnostic method of the present invention specifically comprises the following steps:

[0054] S1. Collect the status data of the oiler device, including normal operation, flow meter failure, inverter failure, metering pump failure, and low oil tank level.

[0055] S2. Collect variable data of the oiling machine device, including inverter output current, inverter fault information, actual speed of the metering pump, temperature of the metering pump, mailbox liquid level, flow meter flow, unit strip speed, and oiling set value.

[0056] S3. Feature extraction specifically includes:

[0057] The data collected in step S1 and step S2 are subjected to feature extraction using a hidden Markov model, and corresponding features are designed according to the operating status of the oiling machine device.

[0058] S4. Training the hidden Markov model; specifically including:

[0059] The hidden Markov model is trained to extract features and estimate the parameters of the hidden Markov model.

[0060] The parameters of the hidden Markov model include:

[0061] The initial state probability vector PI contains the probability of the initial state of the system. For example, [0.3, 0.2, 0.2, 0.2, 0.1] means that at t = 0, the probability of the system being in the "normal operation" state is 0.3, and the probabilities of being in other states are 0.2, 0.2, 0.2 and 0.1 respectively.

[0062] PI = [0.3, 0.2, 0.2, 0.2, 0.1]

[0063] The transition probability matrix contains the transition probabilities between various states. For example, the first row and second column indicate that the probability of transitioning from the "normal operation" state to the "fuel tank level is too low" state is 0.2.

[0064] A=[0.9,0.1,0.0,0.0,0.0]

[0065] [0.2,0.6,0.2,0.0,0.0]

[0066] [0.0,0.3,0.4,0.3,0.0]

[0067] [0.0,0.0,0.4,0.4,0.2]

[0068] [0.0,0.0,0.0,0.1,0.9]

[0069] The observation probability matrix contains the probability distribution of the observed data in each state. For example, the third column of the second row indicates that the probability of observing 400 in the "fuel tank level is too low" state is 0.2.

[0070] B=[0.8,0.1,0.1,0.0,0.0]

[0071] [0.0,0.7,0.2,0.1,0.0]

[0072] [0.0,0.0,0.5,0.4,0.1]

[0073] [0.0,0.0,0.1,0.4,0.5]

[0074] [0.0,0.0,0.0,0.1,0.9]

[0075] S5. Real-time prediction of the state of the oiling machine device specifically includes:

[0076] The Viterbi algorithm is used to infer the most likely state sequence based on the given probability matrix. When an abnormality is detected, the corresponding state sequence is output. The specific steps are as follows:

[0077] 1) Initialization: For the first observation, calculate the probability of all state paths from each possible starting state to the observation, and select the one with the highest probability as the initial state. Calculate the probability of each state appearing at t = 0, which is equal to the product of the initial state probability vector PI and the first value of the observation data. For example, at t = 0, the probability of the "normal operation" state appearing needs to be calculated, that is, 0.3*0.2=0.06.

[0078] 2) Recursion: For each subsequent observation, calculate each possible state as the current state and find a path with the highest probability among the possible states at the previous moment. This way, the most likely path to reach the state and its corresponding probability are obtained.

[0079] 3) Backtracking: After processing all observations, find the path with the highest probability from all possible states as the final result.

[0080] S6. Troubleshooting of the oiling machine device specifically includes:

[0081] Confirm the fault location, analyze the cause of the fault and record it. Confirm the fault location found, analyze the cause, find ways to repair or adjust it, and record it in the database.

[0082] S7. Continuous optimization specifically includes:

[0083] Improve the parameters of the hidden Markov model, update and improve the training data of the hidden Markov model based on data feedback such as the cause of the fault and component loss, and continuously adjust the parameters of the hidden Markov model.

[0084] Example

[0085] This embodiment obtains a set of observation data as follows:

[0086] 1) The inverter output current is 0.2A

[0087] 2) The inverter fault information is 0

[0088] 3) The actual speed of the metering pump is 400rpm

[0089] 4) The metering pump temperature is 60°C

[0090] 5) The tank level is 50%

[0091] 6) The flow rate of the pipe clamp flow meter is 10L / min

[0092] 7) The strip speed of the unit is 300m / min

[0093] 8) Oiling setting value is 1

[0094] Assuming that we want to perform state inference on the above observation data, the specific steps are given below:

[0095] (1) Initialization

[0096] Initialize the value of the first column, that is, the probability at time t = 0. Assume that the current time is 0 and the observed variables are [0.2, 0, 400, 60, 50, 10, 300, 1].

[0097] (2) Recursion

[0098] For each time t, recursively start from t=1 to t=T (T is the time step number of the observed data).

[0099] For each state i, calculate the maximum probability path and the corresponding maximum probability value that appear at time t.

[0100] (3) Backtracking

[0101] At all times, find the state sequence corresponding to the path with the largest total probability value. After performing the calculation, the program will print out the most likely state sequence:

[0102] [0,0,2,3,3,1,0,0]

[0103] The numbers here represent various states, with 0 representing the "normal operation" state, 1 representing the "tank level is too low" state, and so on. Therefore, according to the calculation results, in the 4th and 5th time steps, the machine is most likely in the "tube clamp flow meter failure" state.

[0104] Those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention, and are not intended to limit the present invention. As long as they are within the spirit of the present invention, any changes or modifications to the above embodiments will fall within the scope of the claims of the present invention.

Claims

1. A method for oiler diagnosis based on hidden Markov model combined with Viterbi algorithm, characterized in that: By establishing the corresponding state transfer matrix and observation matrix, a hidden Markov model is established for the operation of the oiler device. The hidden Markov model is trained according to historical data. The observation values ​​generated under different states are assigned to a specific state, and the transition probability between each state is established. The Viterbi algorithm is used to find the most likely state sequence.

2. The oiler diagnosis method based on hidden Markov model combined with Viterbi algorithm according to claim 1 is characterized in that: The oiler diagnostic method specifically comprises the following steps: S1, collecting status data of the oiling machine device; S2, collecting variable data of the oil coating device; S3, feature extraction; S4, training the hidden Markov model; S5, predicting the state of the oiler device in real time; S6, troubleshooting the oiler device; S7. Continuous optimization.

3. According to the oiler diagnostic method based on hidden Markov model combined with Viterbi algorithm in claim 2: the status data in step S1 includes normal operation, flow meter failure, inverter failure, metering pump failure, and low oil tank level.

4. According to the oiler diagnostic method based on hidden Markov model combined with Viterbi algorithm in claim 3: the variable data in step S2 includes inverter output current, inverter fault information, actual speed of metering pump, metering pump temperature, mailbox liquid level, flowmeter flow, unit strip speed, and oiling set value.

5. According to the oiler diagnosis method based on hidden Markov model combined with Viterbi algorithm in claim 4: the feature extraction in step S3 specifically comprises: The hidden Markov model is used to extract features from the data collected in step S1 and step S2, and corresponding features are designed according to the operating status of the oiler device.

6. According to the oiler diagnosis method based on hidden Markov model combined with Viterbi algorithm in claim 5: the step S4 of training the hidden Markov model specifically comprises: The hidden Markov model is used to train and extract features, and the parameters of the hidden Markov model are estimated.

7. According to the oiler diagnosis method based on hidden Markov model combined with Viterbi algorithm in claim 6: the parameters of the hidden Markov model include initial state probability vector PI, transition probability matrix, and observation probability matrix.

8. According to the oiler diagnosis method based on hidden Markov model combined with Viterbi algorithm in claim 7, the step S5 of real-time prediction of the state of the oiler device specifically comprises: The most likely state sequence is inferred according to a given probability matrix by the Viterbi algorithm, and when an abnormality is detected, the corresponding state sequence is output.

9. According to the oiler diagnostic method based on hidden Markov model combined with Viterbi algorithm in claim 8, the step S6 of troubleshooting the oiler device specifically comprises: Confirm the fault location, analyze the cause of the fault and record it. Confirm the fault location found, analyze the cause, find ways to repair or adjust it, and record it in the database.

10. According to the oiler diagnosis method based on hidden Markov model combined with Viterbi algorithm in claim 9, the continuous optimization in step S7 specifically comprises: Improve the parameters of the hidden Markov model, update and improve the training data of the hidden Markov model according to the data feedback such as the cause of the fault and the loss of components, and continuously adjust the parameters of the hidden Markov model.

Images