A fault feature extraction method for flexible hands used in deep-sea submersibles

Abstract

The present invention discloses a fault feature extraction method for a flexible hand for a deep-sea submersible, which belongs to the technical field of flexible hands for submersibles. In order to expand the search range, a new lens imaging inverse learning strategy is integrated in the population initialization; in order to enhance the local search capability of the sparrow-jellyfish hybrid algorithm, a mutation mechanism is introduced in the local development stage of the jellyfish population. The hybrid algorithm will effectively expand the search range, improve the search capability, and avoid falling into the problem of local optimality, so as to efficiently find the optimal combination of penalty factor and decomposition number. The hybrid algorithm will effectively improve the accuracy of fault feature extraction and help improve the reliability of fault diagnosis.

Description

Technical Field

[0001] The invention belongs to the technical field of flexible hands for submersibles, and in particular relates to a method for extracting fault features of flexible hands for deep-sea submersibles. Background Art

[0002] As human beings continue to explore and develop the ocean, deep-sea submersibles have become an important tool. In tasks such as deep-sea resource exploitation, marine environmental monitoring, and biodiversity research, submersible operation manipulators play a key role. Traditional rigid manipulators have exposed many limitations in this environment, such as dynamic sealing problems, weak deep-water pressure resistance, and poor operational flexibility. In order to adapt to the harsh conditions of the deep-sea environment and better complete various delicate operation tasks, flexible manipulators were born.

[0003] The deep-sea environment has extremely high pressure, low temperature, darkness, and complex terrain and water flow conditions, which puts strict requirements on operating equipment. Flexible manipulators can better adapt to these extreme environments. Their flexible structure can operate in narrow spaces and reduce the risk of collision with the surrounding environment. At the same time, flexible manipulators can better withstand deep-sea pressure and reduce the risk of damage that may occur due to rigid structures under high pressure. It can complete various delicate operating tasks, such as seabed resource sampling, equipment maintenance, scientific exploration, etc., providing strong technical support for deep-sea scientific research and resource development.

[0004] Flexible manipulators are generally made of soft or superelastic materials and have many advantages over rigid manipulators. They are more flexible and compliant, can move more flexibly in a limited space, and perform complex operations; because they are made of soft materials, they are lighter and are more suitable for operation in deep-sea environments.

[0005] Deep-sea operations often require submersibles to work independently on the seabed for a long time, which requires the manipulator to have high reliability and autonomy. Any failure may cause mission failure and even cause property loss and environmental damage. Therefore, the exploration of fault diagnosis and related methods for flexible manipulators can effectively improve the safety and efficiency of deep-sea operations, extend the service life of manipulators, and reduce maintenance and repair costs.

[0006] However, fault diagnosis is often inseparable from accurate fault feature extraction. The fault manifestations of flexible manipulators may be complex and hidden. Traditional feature extraction methods may not be able to fully capture the subtle features of the fault, while improved methods can more effectively extract fault-related feature information from various signals (such as current, torque, angular displacement, etc.), thereby more accurately identifying the type, location and severity of the fault. By extracting fault features more accurately, the probability of misjudgment and missed judgment in fault diagnosis can be reduced. Misjudgment may lead to unnecessary maintenance and downtime, increasing costs; missed judgment may prevent the fault from being handled in a timely manner, further exacerbating the damage to the manipulator, and even leading to mission failure. The improved feature extraction method can better adapt to the complex working environment and diverse fault modes of the flexible manipulator and improve the reliability of diagnosis.

[0007] The improvement of fault feature extraction methods is an important driving force for the continuous development of flexible manipulator fault diagnosis technology. By continuously exploring new feature extraction methods and technologies, we can promote technological innovation in the entire fault diagnosis field and provide better technical support for the intelligent and efficient operation of flexible manipulators. Summary of the invention

[0008] In view of the above problems existing in the prior art, the present invention proposes a fault feature extraction method for a flexible hand for a deep-sea submersible, which has a reasonable design, solves the shortcomings of the prior art and has good effects.

[0009] A method for extracting fault features of a flexible hand for a deep-sea submersible comprises the following steps:

[0010] Step 1: Collect historical fault data and real-time data collected by sensors on the flexible hand. The sensors include encoders, piezoelectric acceleration sensors, tension sensors, and current sensors in motor drivers. The encoder collects the angular displacement of the flexible hand, the piezoelectric acceleration sensor collects the acceleration of the flexible arm bar, the tension sensor collects the rope tension, and the current sensor collects the input current of the motor.

[0011] Step 2: Preprocess the data in step 1, including removing outliers, removing noise, and zeroing the mean;

[0012] Step 3: Use the Sparrow-Jellyfish hybrid algorithm to optimize the penalty factor in the variational mode decomposition algorithm and decomposition quantity ;

[0013] Step 4: Decompose the preprocessed data into multiple intrinsic mode functions using the optimized variational mode decomposition;

[0014] Step 5: Extract fault features from the decomposed intrinsic mode function.

[0015] Furthermore, the step 3 includes the following sub-steps:

[0016] Step 3.1: Define the initial values ​​of the penalty factor and the number of decompositions as and ;

[0017] Step 3.2: Use a pseudo-random generator to generate the initial population of jellyfish according to the seed value, define the maximum number of iterations T and the number of jellyfish N, and the position of each jellyfish represents a combination of parameters to be optimized, ;

[0018] Step 3.3: Use the lens imaging inverse learning algorithm to initialize the initial jellyfish population, and set the current iteration number t=1;

[0019] Step 3.4: Start the iteration and perform variational mode decomposition on the preprocessed data according to the initial position of a random jellyfish. Intrinsic mode functions IMF, calculate the envelope entropy of each IMF, and get the average envelope entropy, that is, the fitness value;

[0020] Step 3.5: Then traverse all the remaining jellyfish positions in turn, obtain the average envelope entropy value corresponding to each jellyfish, and select the position corresponding to the jellyfish with the smallest envelope entropy value as the current global optimal position;

[0021] Step 3.6: Update the position of each jellyfish according to the optimal position and time control function, and update the inertia weight in the position update formula , ;

[0022] Step 3.7: The number of iterations is increased by 1. At the tth iteration, the preprocessed data is randomly subjected to variational mode decomposition according to the updated position of a jellyfish in the t-1th iteration, and the average envelope entropy value is calculated. Steps 3.5 to 3.7 are repeated until the maximum number of iterations is reached. At this time, the final parameter combination is obtained. , thus obtaining the variational mode decomposition method with the best decomposition performance.

[0023] Furthermore, in step 3.3, the y-axis is used as a convex lens to model the graph. is the search range of the solution; there is a height of The projection of point P on the x-axis is ; The other side of P is imaged by the lens as an inverted stereoscopic image , whose height is , the projection on the x-axis is , the mapping method is as follows:

[0024] ; (7)

[0025] set up is the scale factor of the lens, we get:

[0026] ; (8)

[0027] in, and They are the first The initial ordinate and abscissa values ​​of the jellyfish. For the Initial horizontal coordinate value of the jellyfish.

[0028] Furthermore, in step 3.4, the objective function of the sparrow-jellyfish hybrid algorithm is the average envelope entropy, and the envelope entropy of each IMF is calculated and recorded as , No. Envelope Entropy of IMF for:

[0029] ; (9)

[0030] in, , To make the envelope The number of intervals obtained by discretizing the value range of , the preprocessed data signal x(t) is obtained by Hilbert transform , which is the envelope; yes The regularized result of The probability distribution under

[0031] Objective Function The expression is:

[0032] ; (10)

[0033] In the formula, is the number of IMF components obtained by variational mode decomposition;

[0034] Traverse all the remaining jellyfish positions in turn to obtain the average envelope entropy value corresponding to each jellyfish, and select the position corresponding to the jellyfish with the smallest envelope entropy value as the current global optimal position.

[0035] Furthermore, the step 3.6 includes the following sub-steps:

[0036] Step 3.61: Calculate the time control function value ;

[0037] Step 3.62: Judgement Is it greater than the constant co? If so, it means that the jellyfish is performing passive motion, i.e., type B motion, and execute step 3.63; otherwise, execute step 3.64;

[0038] Step 3.63: Use the sparrow position update formula to calculate the position update of the jellyfish during passive motion;

[0039] Step 3.64: Judgement Is it less than co? If so, go to step 3.65; otherwise, go to step 3.66;

[0040] Step 3.65: Calculate and compare the fitness values ​​of the mutant jellyfish individuals based on the Gauss-Cauchy mutation mechanism, and update their positions;

[0041] Step 3.66: The jellyfish performs active movement, i.e. type B movement;

[0042] Step 3.67: Determine whether all jellyfish are traversed and their positions are updated. If so, execute step 4. Otherwise, return to execute steps 3.61~3.67.

[0043] Furthermore, in step 3.61, the time control mechanism includes a time control function and a constant co, when When the value of exceeds co, the jellyfish will follow the ocean current; when When the value of is less than co, the jellyfish will enter the jellyfish group; set co to 0.5;

[0044] The expression of the time control function value is:

[0045] ; (11)

[0046] in, The number of the current iteration.

[0047] Furthermore, in step 3.63, the sparrow position update formula is:

[0048] ; (12)

[0049] in, is the position of the i-th sparrow at the t+1th iteration, is the tth iteration The position of a sparrow, is the global best position in the previous iteration; is the step size control parameter; is a random number, and ; and are the current global best and worst fitness values ​​respectively; For the The fitness value of a sparrow, is the fitness value corresponding to the sparrow with the best global position at present, is the position of the sparrow with the worst position in the current global situation, It is an artificially defined constant;

[0050] , The parameter update formula is as follows:

[0051] ; (13)

[0052] ; (14)

[0053] in, = -1.5, = 1.5, = -0.9, =0.9.

[0054] Furthermore, in step 3.65, for Gaussian distribution, its cumulative distribution function is:

[0055] ; (18)

[0056] For the Cauchy distribution, the cumulative distribution function is:

[0057] ; (19)

[0058] Then the cumulative distribution function of the Gaussian-Cauchy mixture distribution is , is the mixing coefficient in [0,1];

[0059] Generate two random numbers on [0,1] and , and The distribution within the value interval is uniform;

[0060] Calculate the mutation value by inversely solving the cumulative distribution function and , that is, solving and Get mutation value and , the mutation vector generated by the Gauss-Cauchy mutation mechanism ;

[0061] After generating the mutation vector, the new position of the jellyfish is:

[0062] ; (20)

[0063] in Represents the tth iteration The new position of the jellyfish after mutation, Represents the tth iteration Only the position of the jellyfish before mutation;

[0064] The formula for updating the jellyfish position after mutation is as follows:

[0065] ;(twenty one)

[0066] in, is the initial position of the jellyfish at the beginning of the t+1th iteration; Representative The fitness value of the position after the jellyfish mutation; Representative The fitness value of the position before the jellyfish mutation;

[0067] In the jellyfish position update phase, for each jellyfish, the probability p is used to decide whether to perform a mutation operation. The probability p is set manually. If a mutation operation is performed, the first Mutation vector of jellyfish ;

[0068] Check the boundary conditions. Due to position mutation, the jellyfish position may move outside the bounded search area. Formula (22) is used to return the jellyfish to the area.

[0069] ;(twenty two)

[0070] In the formula, For the The jellyfish in The location of the dimension; is the updated position after applying the bounding constraints, and The search space is The upper and lower bounds of the dimension;

[0071] Then, the objective function in step 3.4 is used to calculate the fitness value of the jellyfish individual after the mutation, and the fitness values ​​before and after the mutation are compared. If the fitness value after the mutation is better, the position after the mutation is saved as the current position, otherwise the position before the mutation is saved as the current position.

[0072] Further, in step 3.66, randomly select The only one other than jellyfish jellyfish, equations (23) to (25) simulate the Only the jellyfish's step length, movement method, and updated position;

[0073] ;(twenty three)

[0074] ;(twenty four)

[0075] ; (25)

[0076] in, is the step length of the jellyfish, For the Only jellyfish to The vector direction of the jellyfish, For the The only one besides the jellyfish Only the location of the jellyfish, For the Only the location of the jellyfish, For the Only the updated position of the jellyfish, is the fitness value of position X.

[0077] Beneficial technical effects brought by the present invention:

[0078] The present invention proposes a fault feature extraction method for a flexible hand for deep-sea submersibles. In order to expand the search range, a new lens imaging reverse learning strategy is integrated in the population initialization; in order to enhance the local search capability of the hybrid algorithm, a mutation mechanism is introduced in the local development stage of the jellyfish population. The hybrid algorithm will effectively expand the search range, improve the search capability, and avoid falling into the problem of local optimality, so as to efficiently find the optimal combination of penalty factor α and decomposition number k. The hybrid algorithm will effectively improve the accuracy of fault feature extraction and help improve the reliability of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 The present invention is a flowchart of a method for extracting fault features of a flexible hand for a deep-sea submersible.

[0080] Figure 2 It is a schematic diagram of the structure of the flexible hand in the present invention.

[0081] Figure 3 Schematic diagram of the hardware system of the fault feature extraction system in the present invention. DETAILED DESCRIPTION

[0082] The specific implementation of the present invention is further described below in conjunction with specific embodiments:

[0083] A fault feature extraction method for a flexible hand used in deep-sea submersibles, such as Figure 1As shown, the following steps are included:

[0084] Step 1: Collect historical fault data and real-time data collected by sensors on the flexible hand. The sensors include encoders, piezoelectric acceleration sensors, tension sensors, and current sensors in motor drivers. The encoder collects the angular displacement of the flexible hand, the piezoelectric acceleration sensor collects the acceleration of the flexible arm bar, the tension sensor collects the rope tension, and the current sensor collects the input current of the motor.

[0085] The data is divided into historical data and real-time data. Historical data are mostly various fault data, which are used to train fault feature extraction models. Real-time data are various signals collected by various sensors installed on the flexible hand when the flexible hand is working. They are used to input the variational mode decomposition with the best decomposition performance for fault signal decomposition and subsequent fault feature extraction. Historical fault data are mainly divided into the following four categories: motor overload, position deviation, brake failure and speed fluctuation. Real-time data collects current signals, angular displacement signals, rope tension signals and angular acceleration signals under normal working of the flexible hand. The structure of the flexible hand and the installation position of the sensors are as follows: Figure 2 shown.

[0086] The flexible hand is driven by rope. The rotation of the motor connected to the screw motion mechanism causes the screw motion structure to produce linear motion, driving the rope lock arranged on the screw motion structure to move, and then controlling the retraction and release of the driving rope. A tension sensor is arranged on the rope at the rigid connection part of the flexible hand to collect the internal tension data of the rope. To measure the current of the motor, the motor driver has a built-in current sensor, and the current value of the motor can be read through the communication interface of the motor driver. At the joint, the encoder shaft and the joint shaft are connected by a coupling to measure the joint rotation angle to ensure the correct movement of the flexible hand. The high-precision piezoelectric acceleration sensor is installed on the arm of the flexible hand, which can directly measure the acceleration generated by the joint movement to ensure the stability of the flexible hand during operation.

[0087] Step 2: Preprocess the data in step 1, including removing outliers, removing noise, and zeroing the mean;

[0088] The box plot method is used to remove outliers, which includes the minimum value, the first quartile , median, third quartile and maximum values. is the value that separates less than or equal to 25% of the total data. It is the value that divides less than or equal to 75% of the total amount of data. arrive The total box length becomes the interquartile range of the box plot According to the interquartile range To determine outliers. Data points smaller than or greater than The data are considered outliers.

[0089] Wavelet denoising is used to remove noise from the signal. Wavelet denoising is mainly divided into three parts:

[0090] (1) Perform wavelet decomposition on the collected data, and the signal model is expressed as:

[0091] ; (1)

[0092] (2) Wavelet coefficient thresholding. Its equation is expressed as:

[0093] ; (2)

[0094] when When it approaches infinity, the signal noise can be almost completely removed.

[0095] (3) Assign a threshold according to the modified detail coefficient. Compared with hard thresholding, soft thresholding provides smoother results. Equation (2) is also solved by the soft thresholding method, which can achieve good denoising effect. The soft threshold is expressed as follows:

[0096] ; (3)

[0097] (4) Signal reconstruction after inverse transformation of the processed wavelet coefficients. Combining formula (1), we get the following formula:

[0098] ; (4)

[0099] in, is the denoised signal, is the wavelet coefficient;

[0100] Zeroing the mean of the collected data helps to more clearly understand the relationship between data features and target variables, eliminates the influence of the mean, and makes the changes in features more significant. The main steps are as follows:

[0101] (1) Calculate the mean;

[0102] Traverse the read data and calculate the average value of all data points. As follows:

[0103] ; (5)

[0104] (2) Data adjustment;

[0105] For each data point , minus the calculated mean , get the data points after the mean is zeroed , the calculation formula is:

[0106] (6).

[0107] Step 3: Use the Sparrow-Jellyfish hybrid algorithm to optimize the penalty factor in the variational mode decomposition algorithm and decomposition quantity ;

[0108] Variational mode decomposition is an algorithm used in signal processing and data analysis. The input signal f(t) is decomposed into multiple intrinsic mode functions (IMFs), where each IMF component after decomposition is a component of different frequency, which is used for frequency modulation and amplitude modulation. By adding exponential terms, a variational constraint problem is obtained.

[0109] Penalty Factor In the variational mode decomposition algorithm, it is mainly used to control the bandwidth of each decomposed modal function (IMF). A larger penalty factor will make the bandwidth of each IMF narrower, that is, the frequency components in each IMF are more concentrated. This helps to better separate the signal components of different frequency scales, but it may also lead to over-constraint, making the decomposition result too smooth and losing some detailed information. On the contrary, a smaller penalty factor will make the bandwidth of the IMF wider, allowing more components of different frequencies to exist in one IMF. This may make the decomposition result closer to the original signal, but it may also lead to a mixture of different frequency components, making it difficult to clearly separate the signal characteristics of each frequency scale.

[0110] Decomposition quantity It directly determines how many IMFs the original signal is decomposed into. A larger number of decompositions will decompose the original signal more finely, and each IMF may represent a signal component in a narrower frequency range. This may be beneficial for the analysis of complex signals and can better reveal the different frequency characteristics in the signal. However, too large a number of decompositions may also lead to over-decomposition, produce some meaningless IMFs, increase computational complexity, and may introduce false frequency components. A smaller number of decompositions will decompose the signal relatively coarsely, and each IMF may contain signal components in a wider frequency range. This can simplify the analysis in some cases, but may ignore some important frequency features.

[0111] Therefore, the penalty factor and decomposition quantity It plays a key role in variational mode decomposition. The present invention uses a hybrid algorithm of the sparrow-jellyfish search algorithm to reasonably select the penalty factor and the number of decompositions to find the most suitable parameter combination. . In order to obtain accurate and effective signal decomposition results.

[0112] Step 3 includes the following sub-steps:

[0113] Step 3.1: Define the initial values ​​of the penalty factor and the number of decompositions as and ;

[0114] Step 3.2: Use a pseudo-random generator to generate a series of random numbers according to the algorithm and seed value. These numbers are evenly distributed within a certain range. Based on this, the initial population of jellyfish is generated. The maximum number of iterations T and the number of jellyfish N are defined. represents a parameter combination to be optimized. ;

[0115] Step 3.3: Use the lens imaging reverse learning algorithm to initialize the initial jellyfish population, and set the current iteration number t=1;

[0116] The lens imaging reverse learning algorithm is an improved method that can expand the search range by calculating the inverse solution of the current position. This method can expand the scope of the global search and achieve an overall improvement in the quality of individuals in the initial population.

[0117] In step 3.3, the y-axis is used as a convex lens to model the graph. is the search range of the solution; there is a height of The projection of point P on the x-axis is ; The other side of P is imaged by the lens as an inverted stereoscopic image , whose height is , the projection on the x-axis is , the mapping method is as follows:

[0118] ; (7)

[0119] set up is the scale factor of the lens, we get:

[0120] ; (8)

[0121] in, and They are the first The initial ordinate and abscissa values ​​of the jellyfish. For the Initial horizontal coordinate value of the jellyfish.

[0122] Step 3.4: Start the iteration and perform variational mode decomposition on the preprocessed data according to the initial position of a random jellyfish. Intrinsic mode functions IMF, calculate the envelope entropy of each IMF, and get the average envelope entropy, that is, the fitness value;

[0123] In step 3.4, the objective function of the sparrow-jellyfish hybrid algorithm is the mean envelope entropy (MEE), which is calculated for each IMF and recorded as , No. Envelope Entropy of IMF for:

[0124] ; (9)

[0125] in, , To make the envelope The number of intervals obtained by discretizing the value range of , the preprocessed data signal x(t) is obtained by Hilbert transform , which is the envelope; yes The regularized result of The probability distribution under

[0126] Objective Function The expression is:

[0127] ; (10)

[0128] In the formula, is the number of IMF components obtained by variational mode decomposition; obviously, when the value of MME is smaller, the complexity of IMFs is lower and the signal is more stable. Therefore, the minimum MME is selected as the optimization objective function of the hybrid algorithm.

[0129] Traverse all the remaining jellyfish positions in turn to obtain the average envelope entropy value corresponding to each jellyfish, and select the position corresponding to the jellyfish with the smallest envelope entropy value as the current global optimal position.

[0130] Step 3.5: Then traverse all the remaining jellyfish positions in turn, obtain the average envelope entropy value corresponding to each jellyfish, and select the position corresponding to the jellyfish with the smallest envelope entropy value as the current global optimal position.

[0131] Step 3.6: Update the position of each jellyfish according to the optimal position and time control function, and update the inertia weight in the position update formula , ;

[0132] Step 3.6 includes the following sub-steps:

[0133] Step 3.61: Calculate the time control function value c(t);

[0134] The jellyfish search algorithm is mainly based on two phases - exploration and exploitation. In the jellyfish search algorithm, movement toward the ocean current is exploration, movement within the jellyfish group is exploitation, and the time control mechanism switches between them. Initially, the possibility of exploration is greater than the possibility of exploitation to find the area with the best hope of finding food; over time, the possibility of exploitation is much higher than the possibility of exploration, and the jellyfish determines the best position in the search area. Jellyfish either move in the group, defined as passive movement (type A movement), or follow the ocean current, defined as active movement (type B movement); the movement of jellyfish in the jellyfish group is divided into type A (passive movement) and type B (active movement), and the jellyfish switch between these two types of movement. Type A is favored at the beginning; as time goes by, type B becomes more and more favored. In order to simulate this situation, the time control mechanism is introduced. The time control mechanism controls the switching between these types of movement.

[0135] The time control mechanism includes a time control function and a constant co, when When the value of exceeds co, the jellyfish will follow the ocean current; when When the value of is less than co, the jellyfish will enter the jellyfish group; the exact value of co is unknown, and the time control changes randomly from 0 to 1, so co is set to 0.5, which is the average of 0 and 1;

[0136] The expression of the time control function value is:

[0137] ; (11)

[0138] in, The number of the current iteration.

[0139] Used to simulate movement within a swarm of jellyfish (type A or type B). Exceed When , the jellyfish exhibits type A motion. Less than When the jellyfish moves, it shows type B motion. increases from 0 to 1 over time, so The probability initially exceeds Therefore, sports of type A are better than sports of type B. Over time, Approaching 1, The probability of eventually exceeding Therefore, type B sports are more popular.

[0140] Step 3.62: Judgement Is it greater than the constant co? If so, it means that the jellyfish is performing passive motion, i.e., type B motion, and execute step 3.63; otherwise, execute step 3.64;

[0141] Step 3.63: Use the sparrow position update formula to calculate the position update of the jellyfish during passive motion;

[0142] Since the jellyfish search algorithm has limitations in its development capabilities (i.e., passive motion), it is easy to converge prematurely or fall into a local optimum. The sparrow search algorithm is the opposite in terms of development capabilities, so when the jellyfish is in passive motion, the position update formula is replaced by the sparrow position update formula. The sparrow position update formula is:

[0143] ; (12)

[0144] in, is the t+1th iteration The position of a sparrow, is the tth iteration The position of a sparrow, is the global best position in the previous iteration; is the step size control parameter; is a random number, and ; and are the current global best and worst fitness values ​​respectively; For the The fitness value of a sparrow, is the fitness value corresponding to the sparrow with the best global position at present, is the position of the sparrow with the worst position in the current global situation, It is an artificially defined constant;

[0145] In optimization techniques, inertia weights are often used to adjust the treatment between exploration and exploitation. With low inertia weights, exploitation is efficient and exploration is inefficient. On the other hand, a large inertia weight means low exploitation and high exploration. The transition from exploration to exploitation in the original sparrow search algorithm is fixed and cannot be adjusted. As an alternative, the nonlinear decreasing inertia weight enables the adjustment to focus on exploration or exploitation. Due to the lack of exploration in the sparrow search algorithm, the introduction of nonlinear decreasing inertia weights helps to improve the exploration ability and balance it with the existing exploitation. The time-varying β and K values ​​help to ensure high diversity in global exploration in the early stage and exploitation around the global solution in the later stage. Then, sine and cosine functions are used so that they are nonlinear and complement each other.

[0146] , The parameter update formula is as follows:

[0147] ; (13)

[0148] ; (14)

[0149] in, = -1.5, = 1.5, = -0.9, = 0.9. Since the early stage focuses more on exploration, the cognitive coefficient is the largest and the social coefficient is the smallest. In the later stages of the search process, more emphasis is placed on development. In this case, Varies between 1.5 and -1.5, Varies between -0.9 and 0.9.

[0150] Step 3.64: Judgement Is it less than co? If so, go to step 3.65; otherwise, go to step 3.66;

[0151] Step 3.65: Calculate and compare the fitness values ​​of the mutant jellyfish individuals based on the Gauss-Cauchy mutation mechanism, and update their positions;

[0152] When the hybrid algorithm iterates to the later stage, the jellyfish population will move closer to the currently found optimal individual, that is, group aggregation behavior. This behavior will lead to problems such as insufficient population diversity and premature convergence of the algorithm. Once it falls into the local optimum, it will increase the amount of calculation of the algorithm and even mistakenly fall into the local optimal solution.

[0153] In the process of improving swarm intelligence algorithms, the evolution of biological populations is often simulated to improve the algorithms, and introducing mutation factors in population evolution is one of the commonly used methods. The present invention introduces the Gauss-Cauchy mutation mechanism to enhance the diversity of the population.

[0154] Gaussian distribution is also called normal distribution, denoted by Gauss( , ), its distribution expression is as follows:

[0155] ; (15)

[0156] Among them, the random variable x is a Gaussian distribution, which obeys the mathematical expectation ,variance ,as well as =0 and =0; when the variance of the Gaussian distribution is large, the variance factor generated enhances the overall global search ability, which means that the variance range of individuals is larger and more individuals can reach different positions in the search space for search. When the algorithm is trapped in a local optimum, the Gaussian variation factor will give the algorithm a better escape ability. The variance factor generated by the Gaussian distribution with a smaller variance will give the algorithm better local search ability and can obtain more accurate local extreme values ​​earlier, but this will also cause the algorithm to lose some of its local escape ability and easily cause the algorithm to fall into a local optimum.

[0157] In order to solve the problems associated with Gaussian distribution, the present invention also proposes to use Cauchy distribution, which is expressed as follows:

[0158] ; (16)

[0159] The random variable x∈(−∞,+∞) follows the Cauchy distribution with a scale parameter , position parameter x0. yes = 1 and the standard Cauchy distribution at x0 = 0;

[0160] The Gaussian-Cauchy mixture probability density function is defined as:

[0161] ; (17)

[0162] in is the mixing coefficient in [0,1], by adjusting , , , , , the characteristics of the mixed distribution can be controlled.

[0163] For Gaussian distribution , its cumulative distribution function is , usually there is no simple analytical expression, but it can be approximated by numerical methods.

[0164] For the Cauchy distribution , its cumulative distribution function is:

[0165] ; (18)

[0166] Then the cumulative distribution function of the Gaussian-Cauchy mixture distribution is:

[0167] ; (19)

[0168] Generate two random numbers on [0,1] and , and The distribution within the value interval is uniform;

[0169] Calculate the mutation value by inversely solving the cumulative distribution function and , that is, solving and Get mutation value and , the mutation vector generated by the Gauss-Cauchy mutation mechanism .

[0170] After generating the mutation vector, the new position of the jellyfish is:

[0171] ; (20)

[0172] in Represents the tth iteration The new position of the jellyfish after mutation, Represents the tth iteration Only the position of the jellyfish before mutation.

[0173] The formula for updating the jellyfish position after mutation is as follows:

[0174] ;(twenty one)

[0175] in, is the initial position of the jellyfish at the beginning of the t+1th iteration; Representative The fitness value of the position after the jellyfish mutation; Representative The fitness value of the jellyfish's position before mutation.

[0176] This formula means that when the fitness value of the jellyfish's position after mutation is less than or equal to the fitness value of the jellyfish's position before mutation, the jellyfish's position is updated to the position after mutation; when the fitness value of the jellyfish's position after mutation is greater than the fitness value of the jellyfish's position before mutation, the jellyfish's position remains as the position before mutation.

[0177] In the jellyfish position update phase, for each jellyfish, the probability p is used to decide whether to perform a mutation operation. The probability p is set manually. If a mutation operation is performed, the first Mutation vector of jellyfish ;

[0178] Check the boundary conditions. Due to position mutation, the jellyfish position may move outside the bounded search area. Formula (22) is used to return the jellyfish to the area.

[0179] ;(twenty two)

[0180] In the formula, For the The jellyfish The location of the dimension; is the updated position after applying the bounding constraints, and The search space is The upper and lower bounds of the dimension;

[0181] Then, the objective function in step 3.4 is used to calculate the fitness value of the jellyfish individual after the mutation, and the fitness values ​​before and after the mutation are compared. If the fitness value after the mutation is better, the position after the mutation is saved as the current position, otherwise the position before the mutation is saved as the current position.

[0182] Step 3.66: The jellyfish performs active movement, i.e. type B movement;

[0183] To simulate type B motion, randomly select The only one other than jellyfish jellyfish, and use the Only jellyfish to the selected The vector of the jellyfish determines the direction of movement. The amount of food at the location of the jellyfish exceeds that of the second jellyfish of interest. The latter will move to the former when the amount of food at the location of the jellyfish of interest is less than that at the location of the jellyfish of interest. The amount of food for jellyfish is small. The jellyfish will leave the jellyfish. Therefore, each jellyfish in a group will move to a better position to find food. Equations (23), (24), and (25) respectively simulate the step length, movement direction, and updated position of the jellyfish. This movement is considered to be an effective use of the local search space. The amount of food represents the fitness value of the jellyfish's location.

[0184] Equations (23) to (25) simulate the Only the jellyfish's step length, movement method, and updated position;

[0185] ;(twenty three)

[0186] ;(twenty four)

[0187] ; (25)

[0188] in, is the step length of the jellyfish, For the Only jellyfish to The vector direction of the jellyfish, For the The only one besides the jellyfish Only the location of the jellyfish, For the Only the location of the jellyfish, For the Only the updated position of the jellyfish, It's location The fitness value of .

[0189] Step 3.67: Determine whether all jellyfish are traversed and their positions are updated. If so, execute step 4. Otherwise, return to execute steps 3.61~3.67.

[0190] Step 3.7: The number of iterations is increased by 1. At the tth iteration, the preprocessed data is randomly subjected to variational mode decomposition according to the updated position of a jellyfish in the t-1th iteration, and the average envelope entropy value is calculated. Steps 3.5 to 3.7 are repeated until the maximum number of iterations is reached. At this time, the final parameter combination is obtained. , thus obtaining the variational mode decomposition method with the best decomposition performance.

[0191] Step 4: Decompose the preprocessed data into multiple intrinsic mode functions using the optimized variational mode decomposition.

[0192] Step 5: Extract fault features from the decomposed intrinsic mode function.

[0193] The present invention also provides a fault diagnosis system for a flexible hand for a deep-sea submersible, which adopts the above-mentioned fault feature extraction method for a flexible hand for a deep-sea submersible, and includes a hardware system and a software system;

[0194] like Figure 3 As shown, the hardware system includes a flexible hand, a sensor for collecting motion data of the flexible hand in real time during operation, a data acquisition system connected to the sensor for obtaining sensor data, an underwater acoustic communication system for data transmission, a buoy for transmitting the collected data of the flexible hand to the intelligent operation and maintenance department on the ship, a fault diagnosis system for diagnosing faults of the collected signals and taking corresponding measures, an alarm system and a maintenance system.

[0195] 1) Rope flexible hand: It consists of two joints and two rigidly connected arms. The various movements of the flexible hand can be achieved by loosening and tightening the rope through a motor.

[0196] 2) Sensors: collect the motion data of the flexible hand in real time, including encoders, current sensors, tension sensors and piezoelectric acceleration sensors. The functions of each sensor have been described in step 1 of the specific implementation method.

[0197] 3) Data acquisition system: By connecting to various sensors, the data collected by the sensors in real time is obtained and provided to the fault diagnosis system for analysis and judgment.

[0198] 4) Underwater acoustic communication system: realize real-time and efficient underwater data transmission, and transmit the data in the data acquisition card to the buoy.

[0199] 5) Buoy: Communicates with underwater submersibles through hydroacoustic communication, collects sensor data transmitted by the submersibles, maintains communication with ships at sea through satellites, and transmits the collected data to the intelligent operation and maintenance department on the ship for further fault diagnosis and analysis.

[0200] 6) Radio wave communication: At the transmitting end, the information to be transmitted (such as sound, image, data, etc.) is converted into electrical signals through sensors and other equipment. The modulated radio waves are radiated into space through antennas. Radio waves propagate in space in the form of electromagnetic waves and can be transmitted in free space, atmosphere and other environments. At the receiving end, the antenna receives the radio wave signal, and then demodulates the baseband signal in the high-frequency carrier signal through a demodulator to restore the original information.

[0201] The software system includes data management software, communication software, fault diagnosis software, control system software, and health management and maintenance software.

[0202] 1) Data acquisition system: The data collected by the sensor in real time is obtained through the data acquisition card and input into the host computer.

[0203] 2) Fault diagnosis system: By analyzing the historical data and operation data of the flexible robotic arm, the fault diagnosis system can promptly diagnose the faults and problems of the robotic arm and take corresponding solutions in a timely manner.

[0204] 3) Alarm system: The input end is connected to the output end of the data processing system, and an alarm sounds when a flexible robotic arm failure is detected.

[0205] 4) Maintenance system: When the flexible robotic arm fails, take corresponding maintenance measures to ensure that the flexible robotic arm can continue to work or shut down in time.

[0206] 5) Data management software: This software has the functions of data acquisition, data preprocessing, data storage, data retrieval and data analysis.

[0207] 6) Fault diagnosis software: This software has the functions of fault detection, feature extraction, fault diagnosis, fault classification, decision support and module update.

[0208] 7) Communication software: This software has data transmission, protocol management, interface adaptation and network management functions.

[0209] 8) Control system software: has motion control, action scheduling, feedback correction, visual monitoring and fault handling functions.

[0210] 9) Health management and maintenance software: with real-time monitoring, abnormality detection, safety alarm, automatic response, braking stop and maintenance and inspection reminder functions.

[0211] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A fault feature extraction method for a flexible hand for a deep-sea submersible, characterized in that: The following steps are involved: Step 1: Collect historical fault data and real-time data collected by sensors on the flexible hand, including encoders, piezoelectric accelerometers, tension sensors, and current sensors in motor drivers; The encoder collects the angular displacement of the flexible hand, the piezoelectric acceleration sensor collects the acceleration of the flexible arm rod, the tension sensor collects the rope tension, and the current sensor collects the input current of the motor; Step 2: Preprocess the data in step 1, including removing outliers, removing noise, and zeroing the mean; Step 3: Use the sparrow-jellyfish hybrid algorithm to optimize the penalty factor α and the number of decompositions k in the variational mode decomposition algorithm; Step 4: Decompose the preprocessed data into multiple intrinsic mode functions using the optimized variational mode decomposition; Step 5: Extract fault features from the decomposed intrinsic mode function; The step 3 includes the following sub-steps: Step 3.1: Define the initial values ​​of the penalty factor and the number of decompositions as α0 and k0; Step 3.2: Use a pseudo-random generator to generate the initial population of jellyfish according to the seed value, define the maximum number of iterations T and the number of jellyfish M, and the position X of each jellyfish i represents a parameter combination to be optimized, i = 1, 2, ..., M; Step 3.3: Use the lens imaging inverse learning algorithm to initialize the initial jellyfish population, and set the current iteration number t = 1; Step 3.4: Start iteration, randomly perform variational mode decomposition on the preprocessed data according to the initial position of a jellyfish, decompose to obtain k intrinsic mode functions IMF, calculate the envelope entropy of each IMF, and obtain the average envelope entropy, that is, the fitness value; Step 3.5: Then traverse all the remaining jellyfish positions in turn, obtain the average envelope entropy value corresponding to each jellyfish, and select the position corresponding to the jellyfish with the smallest envelope entropy value as the current global optimal position; Step 3.6: Update the position of each jellyfish according to the optimal position and time control function, and update the inertia weights β and K in the position update formula; Step 3.7: The number of iterations is increased by 1. At the tth iteration, the preprocessed data is randomly subjected to variational mode decomposition according to the updated position of a jellyfish in the t-1th iteration, and the average envelope entropy value is calculated. Steps 3.5 to 3.7 are repeated until the maximum number of iterations is reached. At this time, the final parameter combination [k, α] is obtained, thereby obtaining the variational mode decomposition method with the best decomposition performance.

2. The fault feature extraction method for a flexible hand for a deep-sea submersible according to claim 1 is characterized in that: In step 3.3, the y-axis is used as a convex lens to model the graph, and [a, b] is the search range of the solution; there is a point P with a height of h, and the projection of point P on the x-axis is x; the other side of P is imaged by the lens as an inverted stereo image P * , the projection on the x-axis is x * , the mapping method is as follows: Let m = h / h * is the scale factor of the lens, we get: Among them, h * and are the initial ordinate and abscissa values ​​of the i-th jellyfish after reverse solution, x i is the initial horizontal coordinate value of the i-th jellyfish.

3. The fault feature extraction method for a flexible hand for a deep-sea submersible according to claim 2 is characterized in that: In step 3.4, the objective function of the sparrow-jellyfish hybrid algorithm is the average envelope entropy. The envelope entropy of each IMF is calculated. Let r = 1, 2, ..., k. The envelope entropy E of the rth IMF is Pr for: Where j = 1, 2, ..., N, N is the envelope a r,j The number of intervals obtained by discretizing the range of the value of , the preprocessed data signal x(t) is obtained by Hilbert transform r,j , which is the envelope; p r,j is a r,j The regularized result of r,j The probability distribution under The expression of the objective function MEE is: Where k′ is the number of IMF components obtained by variational mode decomposition; Traverse all the remaining jellyfish positions in turn to obtain the average envelope entropy value corresponding to each jellyfish, and select the position corresponding to the jellyfish with the smallest envelope entropy value as the current global optimal position.

4. The method for extracting fault features of a flexible hand for a deep-sea submersible according to claim 3, characterized in that: The step 3.6 includes the following sub-steps: Step 3.61: Calculate the time control function value c(t); Step 3.62: Determine whether c(t) is greater than the constant co. If so, it means that the jellyfish is performing passive motion, that is, type B motion, and execute step 3.

63. Otherwise, execute step 3.

64. Step 3.63: Use the sparrow position update formula to calculate the position update of the jellyfish during passive motion; Step 3.64: Determine whether 1-c(t) is less than co, if so, execute step 3.65, otherwise execute step 3.66; Step 3.65: Calculate and compare the fitness values ​​of the mutant jellyfish individuals based on the Gauss-Cauchy mutation mechanism, and update their positions; Step 3.66: The jellyfish performs active movement, i.e. type B movement; Step 3.67: Determine whether all jellyfish are traversed and their positions are updated. If so, execute step 4. Otherwise, return to execute steps 3.61 to 3.

67.

5. The fault feature extraction method for a flexible hand for a deep-sea submersible according to claim 4 is characterized in that: In step 3.61, the time control mechanism includes a time control function c(t) and a constant co. When the value of c(t) exceeds co, the jellyfish will follow the ocean current; when the value of c(t) is less than co, the jellyfish will enter the jellyfish group; co is set to 0.5; The expression of the time control function value is: Among them, t is the current iteration number.

6. The method for extracting fault features of a flexible hand for a deep-sea submersible according to claim 5, characterized in that: In step 3.63, the sparrow position update formula is: in, is the position of the i-th sparrow at the t+1th iteration, is the position of the i-th sparrow at the t-th iteration, is the global best position in the previous iteration; β is the step size control parameter; K is a random number, and K∈[-1,1]; f g and f w are the current global best and worst fitness values ​​respectively; f i is the fitness value of the i-th sparrow, f g is the fitness value corresponding to the sparrow with the best global position at present, is the position of the sparrow with the worst position in the current global situation, and ε is an artificially defined constant; The update formula of β and K parameters is as follows: among them,b min =-1.5,β max =1.5, K min =-0.9, K max = 0.

9.

7. The method for extracting fault features of a flexible hand for a deep-sea submersible according to claim 6, characterized in that: In step 3.65, for Gaussian distribution, its cumulative distribution function is: For the Cauchy distribution, the cumulative distribution function is: Then the cumulative distribution function of the Gaussian-Cauchy mixture distribution is F(x) = λF Gauss (x)+(1-λ)F Cauchy (x), λ is the mixing coefficient in [0,1]; Generate two random numbers u1 and u2 on [0,1], and the distribution of u1 and u2 in the value range is uniform; The mutation values ​​Δx1 and Δx2 are calculated by inversely solving the cumulative distribution function, that is, solving F(Δx1)=u1 and F(Δx2=u2 to obtain the mutation values ​​Δx1 and Δx2. The mutation vector ΔX generated by the Gauss-Cauchy mutation mechanism is i =(Δx1, Δx2); After generating the mutation vector, the new position of the jellyfish is: in represents the new position of the ith jellyfish after mutation at the tth iteration, represents the position of the i-th jellyfish before mutation at the t-th iteration; The formula for updating the jellyfish position after mutation is as follows: in, is the initial position of the jellyfish at the beginning of the t+1th iteration; f i,new represents the fitness value of the position of the i-th jellyfish after mutation; f i,old represents the fitness value of the position of the i-th jellyfish before mutation; In the jellyfish position update phase, for each jellyfish, the probability p is used to decide whether to perform a mutation operation, and the probability p is artificially determined; if a mutation operation is performed, the mutation vector ΔX of the i-th jellyfish is generated according to the above Gaussian-Cauchy mixture probability density function i ; Check the boundary conditions. If the jellyfish position moves outside the bounded search area, use formula (22) to return the jellyfish to the area; Where, X i,d is the position of the i-th jellyfish in the d-th dimension; X′ i,d is the updated position after applying the boundary constraints, U b,d and L b,d are the upper and lower bounds of the search space in the dth dimension respectively; Then, the objective function in step 3.4 is used to calculate the fitness value of the jellyfish individual after the mutation, and the fitness values ​​before and after the mutation are compared. If the fitness value after the mutation is better, the position after the mutation is saved as the current position, otherwise the position before the mutation is saved as the current position.

8. The method for extracting fault features of a flexible hand for a deep-sea submersible according to claim 7, characterized in that: In step 3.66, the nth jellyfish other than the i-th jellyfish is randomly selected, and equations (23) to (25) respectively simulate the step length, movement method and updated position of the i-th jellyfish; in, is the step length of the jellyfish, is the vector direction from the i-th jellyfish to the n-th jellyfish, X n (t) is the position of the nth jellyfish except the i-th jellyfish, X i (t) is the position of the i-th jellyfish, X i (t+1) is the updated position of the i-th jellyfish, and f(·) is the fitness value of position X.

Images