A fault diagnosis method for AUV propulsion system

Abstract

The present invention proposes a method for diagnosing AUV thruster faults, firstly estimating the motor load and propeller torque; then using the estimated motor load, propeller torque and motor control signal as input, using an autoencoder to identify the fault; if the reconstruction error of the autoencoder exceeds the threshold, a fault alarm is output, and if the alarm lasts for T0 time, it is determined to be a thruster fault; finally, a fault isolation table is constructed to analyze and determine the cause of the fault. This scheme does not rely on a large amount of fault data when diagnosing and identifying thruster faults. It only relies on the measurement data of the sensors carried by the AUV itself to identify and determine the causes of various faults such as feedback faults of motor current and speed, as well as propeller loss, shedding and circuit open, and has higher practical application value.

Description

Technical Field

[0001] The present invention relates to the technical field of propeller system fault diagnosis, and in particular to a method for diagnosing a fault of an AUV propeller system. Background Art

[0002] Autonomous underwater vehicles (AUVs) are becoming increasingly popular in terms of wide application and acceptance in defense, marine, and industrial applications. As an important tool for ocean exploration, AUVs must be safe and reliable when working underwater. Therefore, fault diagnosis technology has become one of the most important research topics in this field.

[0003] At present, most commercial AUVs mainly use underwater thrusters as actuators when sailing underwater. Therefore, thruster failure is one of the most common sources of failure for AUVs. In fact, once the propulsion system of an AUV fails, not only will the mission fail to be completed, but the AUV itself will also face the risk of being lost or damaged. To avoid this situation, studying a timely and effective fault diagnosis strategy is conducive to reducing the risk of AUV damage and avoiding the deep propagation of faults, which is of great significance to ensuring the safety of AUVs in complex marine environments and improving their maneuverability.

[0004] Since the 1990s, people have conducted a lot of research on propeller fault diagnosis. At present, the methods for propeller fault diagnosis are divided into three categories, including: based on analytical models, based on data-driven and based on hybrid methods. For example, the invention patent with application publication number CN113283292A discloses a method and device for underwater micro-thruster fault diagnosis, including a model training stage, a fault diagnosis stage and a model optimization stage. In the model training stage, multiple groups of motor current historical signals of the propeller under different fault types are collected as the training sample data set of the model, and the improved HHT transformation is used to extract the feature vector, and then the feature vector is extracted based on the extracted feature vector and combined with the underwater micro-thruster fault type to obtain the propeller fault diagnosis model; in the fault diagnosis stage, the real-time signal of the motor current of the propeller is collected to extract its feature vector, and the feature vector is input into the trained fault diagnosis model to determine the fault state of the propeller; in the model optimization stage, the collected data information is added to the pre-established model training sample data set, and combined with the updated propeller fault type for training, an optimized propeller fault diagnosis model is obtained.

[0005] There are two problems with existing propulsion system fault diagnosis methods: 1. Most methods for directly diagnosing thruster faults rely on feedback information from the thrusters. Once the feedback information is wrong, it will lead to misdiagnosis of the fault; 2. Although using the spacecraft model to identify thruster faults has good results, it is impossible to determine the cause of the fault. Summary of the invention

[0006] In order to solve the defects of the existing thruster fault diagnosis scheme, such as poor accuracy and inability to determine the fault cause, the present invention proposes an AUV thruster system fault diagnosis method, which can not only diagnose the thruster fault, but also determine the fault cause.

[0007] The present invention is implemented by adopting the following technical solution: A method for diagnosing faults of an AUV thruster, characterized in that it comprises the following steps:

[0008] Step A, estimating the motor load and propeller torque;

[0009] Step B, using the estimated motor load, propeller torque and motor control signal as input, using an autoencoder to identify the fault;

[0010] If the reconstruction error of the autoencoder exceeds the set threshold, a fault alarm is output. If the alarm lasts for T0 time, it is determined to be a thruster failure;

[0011] Step C: construct a fault isolation table to analyze and determine the cause of the fault: combine the motor control signal to distinguish between thruster current feedback fault, motor speed feedback fault, propeller loss, propeller damage, and open circuit fault.

[0012] Furthermore, in step A, when estimating the motor load, an extended state observer is used, specifically:

[0013] (1) Construct the motor motion equation:

[0014]

[0015] Among them, T e is the motor electromagnetic torque, Q m is the load torque of the motor, J M is the motor moment of inertia, B v is the motor friction coefficient, n is the motor speed;

[0016] (2) Estimate the motor load and the extended state observer is expressed as:

[0017]

[0018] in, is an estimate of n, is the estimated value of λ, k1 and k2 are the observation gains designed for the extended state observer. When the observer converges, according to Get the observed motor load.

[0019] Furthermore, in step A, the propeller torque is estimated using the propeller torque equation:

[0020] (1) The torque equation of the propeller is expressed as:

[0021]

[0022] Where g() is the torque polynomial, θ and T θ is the change of tilt angle and period caused by pitch motion, ψ and T ψ is the change in tilt angle and period due to yaw motion, and h is the AUV depth change;

[0023] (2) Estimation of propeller torque based on neural network:

[0024] u,θ,ψ,h,n p Five parameters are used as inputs to the neural network, where u is the speed of the AUV equipment, n p is the propeller speed, and the output of the neural network is the propeller torque.

[0025] Furthermore, in step B, the motor load estimation Q m , propeller torque estimation Q p and the motor control signal s, input into the sparse autoencoder, and its error is calculated as follows:

[0026]

[0027] The above error is compared with the thresholds r1, r2, and r3 to determine whether a fault has occurred. The fault judgment criteria are as follows:

[0028]

[0029] Among them, r1, r2, and r3 correspond to the motor load estimation Q m , propeller torque estimation Q p and the set threshold of the motor control signal s.

[0030] Further, in the step C, the moment when the fault is first identified and maintained for T0 seconds is defined as the fault identification point; the fault includes current feedback fault, speed feedback fault, propeller loss, propeller winding and open circuit fault;

[0031] Define the correlation coefficient Cx1,x2 between any two faults x1 and x2:

[0032]

[0033] Where t is the fault identification point, T0 is the observation period, η x1 and η x2 is the average of variables x1 and x2, The value range is -1 to 1;

[0034] Assuming the correlation coefficient |Cx1,x2|>M, there is a correlation between x1 and x2, and the following fault analysis and determination conditions are available:

[0035] Current feedback fault: 0 <c Qm,s ≤M,M≤c Qp,s and 0 <c Qm,Qp ;

[0036] Speed ​​feedback fault: M≤c Qm,s , 0 <c Qp,s ≤M and 0 <c Qm,Qp ;

[0037] Propeller lost: c Qm,s ≤-M,M≤c Qp,s And c Qm,Qp <0;

[0038] Propeller winding: M≤c Qm,s , c Qp,s ≤-M and c Qm,Qp <0;

[0039] Open circuit fault: c Qm,s ≤-M,c Qp,s ≤-M and c Qm,Qp >0;

[0040] Among them, c Qm,s is the correlation coefficient between the motor load and the motor control signal, c Qp,s is the correlation coefficient between propeller torque and motor control signal; c Qm,Qp is the correlation coefficient between motor load and propeller torque.

[0041] Compared with the prior art, the advantages and positive effects of the present invention are:

[0042] This scheme first estimates the motor load based on the current and motor speed, estimates the propeller torque based on the motion state of the AUV and the motor speed, and uses a sparse autoencoder to identify the fault in combination with the motor control signal. After the fault is identified, under the same motor control signal, the motor load and propeller torque will show different change trends according to different fault types, and then use the correlation coefficient between the motor control signal, propeller torque and motor load to accurately determine the cause of the fault.

[0043] This solution does not rely on a large amount of fault data when diagnosing and identifying thruster faults. It only relies on the measurement data of the sensors carried by the AUV itself to identify and determine the causes of various faults such as motor current and speed feedback faults, propeller loss, shedding, and circuit open. It has high monitoring accuracy and higher practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a schematic diagram of the overall process of the fault diagnosis method according to an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of the neural network structure described in an embodiment of the present invention;

[0046] Figure 3 Schematic diagram of the structure of the sparse autoencoder according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention can also be implemented in other ways different from those described herein, and therefore, the present invention is not limited to the specific embodiments disclosed below.

[0048] This embodiment proposes an AUV thruster fault diagnosis method, comprising the following steps:

[0049] Step A, estimating the motor load and propeller torque;

[0050] Step B, using the estimated motor load, propeller torque and motor control signal as input, using an autoencoder to identify the fault;

[0051] If the reconstruction error of the autoencoder exceeds the threshold, a fault alarm is output. If the alarm lasts for T0 time, the reconstruction error exceeds the threshold during this period, and it is determined to be a thruster fault;

[0052] Step C: construct a fault isolation table to analyze and determine the cause of the fault: combine the motor control signal to distinguish between thruster current feedback fault, motor speed feedback fault, propeller loss, propeller damage, and open circuit fault.

[0053] Specifically, Figure 1As shown, this scheme first estimates the motor load and propeller torque. When estimating the motor load, the battery current and motor speed are mainly used, and when estimating the propeller torque, the motion state of the AUV and the motor speed are used. The autoencoder is then used to identify the fault. When a fault occurs, the residual between the input and output of the autoencoder will exceed the set threshold. After the fault is identified, under the same motor control signal, the motor load and propeller torque will show different change trends according to different fault types, and the correlation coefficient between the motor control signal, propeller torque and motor load is used to accurately determine the cause of the fault. In order to more clearly understand the present invention, the present invention is further described below in conjunction with specific embodiments:

[0054] In step A, the motor load and propeller torque are estimated to describe the state of the propeller from different angles. This embodiment mainly uses an extended state observer (ESO) and a neural network (NN) to estimate the motor load and propeller torque respectively:

[0055] 1. First, use the extended state observer to estimate the motor load:

[0056] (1) Construct the motor motion equation, as shown in formula (1):

[0057]

[0058] Among them, T e is the motor electromagnetic torque, Q m is the load torque of the motor, J M is the motor moment of inertia, B v is the motor friction coefficient, n is the motor speed. When the motor is in a stable working state, the motor electromagnetic torque can be represented by the current, the load torque can be equivalent to the propeller, the motor acceleration is 0, and the motor speed is constant.

[0059] Define λ=(B v n+Q m ) / J M As a new variable, rewrite equation (1) into equation (2):

[0060]

[0061] Where Cm is the motor torque coefficient and y is the system output.

[0062] (2) Design an extended state observer to estimate the motor load:

[0063]

[0064] in is an estimate of n, is the estimated value of λ, k1 and k2 are the observation gains designed for the extended state observer. When the observer converges, The observed motor load can be obtained.

[0065] 2. Generally, the torque of the propeller can be approximately represented by the motor load, or can be calculated using the propeller torque equation. In this embodiment, the propeller torque is estimated using the propeller torque equation, which is as follows:

[0066] (1) Considering that the AUV has not only vertical motion but also horizontal motion, when the propeller is far away from the water surface, its vertical displacement has little effect on the performance of the propeller. Therefore, the torque equation of the propeller installed on the AUV is expressed as:

[0067]

[0068] Where g() is the torque polynomial, θ and T θ is the change of tilt angle and period caused by pitch motion, ψ and T ψ is the change of tilt angle and period due to yaw motion, and h is the change of AUV depth. Considering the nonlinearity and uncertainty in g(), this embodiment introduces a neural network (see Figure 2 ) is used to approximate the propeller torque model under complex sea conditions.

[0069] (2) Estimation of propeller torque based on neural network:

[0070] from Figure 2 It can be seen that the neural network contains 5 input variables, 1 hidden layer and 1 output variable. The 5 input variables include: u, θ, ψ, h, n p , u is the speed of the AUV equipment, these parameters can be obtained through sensors or simple calculations. It should be noted that T θ and T ψ It can be calculated by and , so they are not used as the input of the neural network. The output of the neural network is the propeller torque.

[0071] It should be noted that the method of estimating the motor load using the motor state equation, motor current, and motor speed can be implemented by other means, such as neural network, observer, filtering algorithm, etc. Similarly, the method of estimating the propeller torque using the AUV motion state and motor speed can also be implemented by using least squares method, neural network, observer, filtering algorithm, etc., the purpose of which is to observe the motor load and propeller torque. There is no specific limitation on the implementation method, and those skilled in the art can implement it according to common knowledge and experience.

[0072] In step B, considering the influence of the external environment on the propeller torque estimation, this embodiment introduces a sparse autoencoder to identify the fault. At the same time, the fault observation period is set, which can avoid the misidentification of faults caused by large peak reconstruction errors. The sparse autoencoder is a neural network with sparse terms, which is trained to reconstruct its input (see Figure 3 ). The purpose of this step is to reconstruct normal data with lower errors and abnormal data with higher errors while avoiding overfitting, and then apply the size of the error to determine whether a fault exists.

[0073] The sparse autoencoder contains encoder, hidden layer and decoder functions. The encoder maps the input data to the hidden representation. And the decoder reconstructs the hidden layer features. After the sparse autoencoder is trained, based on the reconstruction error, the sequence is classified as normal or abnormal for Q m , Q p and the motor control signal s, the error is calculated as follows:

[0074]

[0075] The residuals generated are different from the thresholds r1, r2, and r3 (r1, r2, and r3 correspond to Q m , Q p The threshold value is compared with the motor control signal s to determine whether a fault has occurred. It should be noted that the threshold value is determined based on engineering experience and can be set based on experience and actual needs. The fault judgment criteria are as follows:

[0076]

[0077] In practical application of AUV, Q m , Q p Both s and s are disturbed by the external environment (for example, when an AUV suddenly changes its course, the ocean current will cause a disturbance different from before), and when the input is reconstructed using a sparse autoencoder, there may be a peak error exceeding the threshold. Therefore, if (6) is directly applied for fault identification, false alarms may occur. For this problem, this embodiment outputs a fault warning and then observes whether the reconstruction error continues to exceed the threshold for a period of time, such as T0 seconds. In this paper, the observation period T0 is set to 30s to determine the fault.

[0078] It should be noted that this embodiment uses a sparse autoencoder as an example for fault diagnosis. In addition to the sparse autoencoder, a denoising autoencoder or the like may also be used. The purpose is to realize fault diagnosis using the motor load, propeller torque and motor control signal as input.

[0079] In step C, once the fault is identified, a key issue is how to determine the cause of the fault, i.e., fault isolation. m Considering that the speed of the AUV is not affected by the current feedback failure, Q p and s still maintain the same trend. Contrary to the current feedback fault, when the motor speed feedback fault occurs, Q p Mutation occurs. At the same time, Q m There will be slight fluctuations, but the impact is limited to system losses. m The trend of s is consistent. For propeller loss, from formula (1), we can know that Q m It should be close to zero. At this point, the speed of the AUV decreases, the control signal gradually increases, and the motor speed increases with the control signal until it reaches the maximum. Then, Q p Increase to the maximum torque. On the contrary, when the propeller winding occurs, the motor speed is zero. As the AUV decelerates, the control signal gradually increases and the motor current increases to the maximum. At this time, it can be seen from formulas (1) and (4) that Q p is zero, Q m will increase to maximum torque. When an open circuit fault occurs, the thruster is disconnected from the battery. Both the current and the motor speed are zero, so Q m and Q p Then, the thruster control signal reaches the maximum value under the regulation of the AUV control system.

[0080] In order to describe these trends, this embodiment introduces a correlation coefficient to calculate the Q around the fault identification point. m With s, Q p With s, Q m With Q p The moment when the fault is first recognized and maintained for 30 seconds is defined as the fault recognition point. The correlation coefficient between two given variables x1 and x2 is defined as follows:

[0081]

[0082] Where t is the fault identification point. T0 is the observation period. η x1 and η x2 is the average of variables x1 and x2. The value range is -1 to 1. When (in this embodiment, M is 0.7, and the M value is determined based on experience and actual needs), it shows that there is a strong positive correlation between x1 and x2. When , it indicates that there is a strong negative correlation between x1 and x2. The above analysis can be converted into symbolic representation and summarized in Table 1.

[0083] Table 1 Fault isolation table

[0084]

[0085] In summary, this method utilizes the motion state information of the AUV, combined with the motor model in the thruster and the control signal of the AUV, to achieve fault identification and also to determine the cause of the fault, and can distinguish between current feedback fault, speed feedback fault, propeller winding, propeller loss and system open circuit fault.

[0086] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for diagnosing AUV thruster faults, characterized in that: The following steps are involved: Step A, estimating the motor load and propeller torque; Step B, using the estimated motor load, propeller torque and motor control signal as input, using an autoencoder to identify the fault; If the reconstruction error of the autoencoder exceeds the set threshold, a fault alarm is output. If the alarm lasts for T0 time, it is determined to be a thruster failure; Step C: construct a fault isolation table to analyze and determine the cause of the fault: combine the motor control signal to distinguish between propeller current feedback fault, motor speed feedback fault, propeller loss, propeller damage, and open circuit fault; The moment when a fault is first identified and maintained for T0 seconds is defined as the fault identification point; the faults include current feedback fault, speed feedback fault, propeller loss, propeller winding and open circuit fault; Define the correlation coefficient Cx1,x2 between any two variables x1 and x2: Where t is the fault identification point, T0 is the observation period, η x1 and η x2 is the average of variables x1 and x2, The value range is -1 to 1; Assuming the correlation coefficient |Cx1,x2|>M, there is a correlation between x1 and x2, and the following fault analysis and determination conditions are available: Current feedback fault: 0 <c Qm,s ≤M,M≤c Qp,s and 0 <c Qm,Qp ; Speed ​​feedback fault: M≤c Qm,s , 0 <c Qp,s ≤M and 0 <c Qm,Qp ; Propeller lost: c Qm,s ≤-M,M≤c Qp,s And c Qm,Qp <0; Propeller winding: M≤c Qm,s , c Qp,s ≤-M and c Qm,Qp <0; Open circuit fault: c Qm,s ≤-M,c Qp,s ≤-M and c Qm,Qp >0; Among them, c Qm,s is the correlation coefficient between the motor load and the motor control signal, c Qp,s is the correlation coefficient between propeller torque and motor control signal; c Qm,Qp is the correlation coefficient between motor load and propeller torque.

2. The AUV thruster fault diagnosis method according to claim 1, characterized in that: In step A, when estimating the motor load, an extended state observer is used, specifically: (1) Construct the motor motion equation: Among them, T e is the motor electromagnetic torque, Q m is the load torque of the motor, J M is the motor moment of inertia, B v is the motor friction coefficient, n is the motor speed; (2) Estimate the motor load and the extended state observer is expressed as: in, is an estimate of n, is the estimated value of λ, λ=(B v n+Q m ) / J M , Cm represents the motor torque coefficient, k1 and k2 are the observation gains designed for the extended state observer. When the observer converges, according to Get the observed motor load.

3. The AUV thruster fault diagnosis method according to claim 2, characterized in that: In step A, the propeller torque is estimated using the propeller torque equation: (1) The torque equation of the propeller is expressed as: Where g() is the torque polynomial, θ and T θ is the change of tilt angle and period caused by pitch motion, ψ and T ψ is the change in tilt angle and period due to yaw motion, and h is the AUV depth change; (2) Estimation of propeller torque based on neural network: With u,θ,ψ,h,n p Five parameters are used as the input of the neural network, where u is the speed of the AUV equipment, n p is the propeller speed, and the output of the neural network is the propeller torque.

4. The AUV thruster fault diagnosis method according to claim 1, characterized in that: In step B, the motor load estimation Q m , propeller torque estimation Q p and the motor control signal s, input into the sparse autoencoder, and its error is calculated as follows: and They correspond to the predicted values ​​of the motor load, propeller torque and motor control signal output by the sparse autoencoder respectively; The above error is compared with the thresholds r1, r2, and r3 to determine whether a fault has occurred. The fault judgment criteria are as follows: Among them, r1, r2, and r3 correspond to the motor load estimation Q m , propeller torque estimation Q p and the set threshold of the motor control signal s.

Images