Method for arranging active noise control sensor actuators in helicopter cabin

Abstract

The invention discloses a method for arranging sensors and actuators for active noise control in a helicopter cabin, which introduces machine learning into a sensor / actuator optimization problem in active noise control in the helicopter cabin, provides a BPNN-GA algorithm, and mainly comprises two processes: sensor optimization based on a neural network and actuator optimization based on a genetic algorithm. Firstly, radiation surface element division is carried out; secondly, the positions and the number of sensors are optimized through machine learning, that is, a high-precision top plate vibration-radiation noise model is established through a neural network, and main sound radiation surface elements are recognized based on the model and serve as arrangement points of the sensors; and finally, an actuator is optimized by using a genetic algorithm, so that the noise level in the cabin is reduced to the greatest extent while the vibration of the control point is effectively suppressed. Compared with a traditional genetic algorithm, the method has the advantages that the calculation efficiency is greatly improved while the control effect is guaranteed, and the method has more obvious advantages in the aspect of improving the calculation efficiency along with the increase of the optimization scale.

Description

Technical Field

[0001] The invention belongs to the field of helicopter cabin noise control sensor / actuator optimization method, and in particular relates to a helicopter cabin noise active control sensor actuator optimization algorithm. Background Art

[0002] The problem of helicopter cabin noise is very serious. Severe cabin noise affects the physical and mental health of the crew members. At the same time, the helicopter structural vibration caused by noise will also affect the performance and life of the helicopter. Among them, the medium and high frequency harmonic noise generated by the meshing vibration of the main reducer gears is the focus of attention. Its frequency range is roughly 500-2000Hz, which is the frequency band range that the human ear is sensitive to. Studies have shown that the vibration of the main reducer will be transmitted to the cabin wall through the supporting structure, excite the wall panel to vibrate, and radiate noise into the cabin. This structural sound transmission is the main source of noise in the helicopter cabin. Therefore, it is very important to reduce the helicopter cabin noise caused by this structural sound transmission.

[0003] Active control technology and passive control technology are two common noise control technologies in the field of helicopter noise control. The essence of passive control technology is to install specific energy dissipation devices at specific locations of the structure. These devices use vibration isolation and vibration absorption technology to convert energy, thereby reducing the energy absorbed by the main structure and achieving the goal of reducing vibration and noise. However, the use of traditional passive control technology to control medium and low frequency noise has a large system size and heavy mass, which will increase the helicopter's fuel consumption and have an adverse effect on the helicopter's performance. In addition, passive control technology is usually unable to adapt to changes in the helicopter's rotation speed, and the noise suppression effect is difficult to achieve as expected.

[0004] Active control technology is divided into active sound cancellation (ASC) and active structural acoustic control (ASAC). Although ASC can solve the problem of low-frequency noise in the cabin, it is impractical to achieve global noise control for the high-frequency noise of gear meshing due to the high modal density of the acoustic cavity. ASAC uses an active control algorithm to control mechanical vibration using actuators installed on the vibration transmission path, and reduces radiated noise by reducing vibration. For example, helicopters such as EH-101, BK-117 and Bell-407 have carried out ASAC tests, and effective control of cabin noise has been achieved through actuators installed on the main struts. However, for helicopters such as S-72 and UH-60 that do not have main struts, ASAC technology based on active struts will no longer be applicable.

[0005] The wall panel is the main transmission path and direct radiation source of noise in the helicopter cabin. Its acoustic and vibration characteristics directly affect the cabin acoustic environment. The main reducer is directly connected to the top panel through the support structure. The top panel is the key forced vibration area and the main source of radiated noise. Reducing its vibration level can effectively reduce radiated noise and prevent vibration energy from being transmitted to other wall panels in the cabin. With the continuous development of smart material technology, ASAC technology based on smart active wall panels has become the most valuable and promising helicopter cabin noise control technology. Summary of the invention

[0006] In view of the above-mentioned defects, the present invention provides a method for deploying sensor actuators for active noise control in a helicopter cabin, which integrates the actuators, sensors and the helicopter roof to form an intelligent active wall panel, which can sense the vibration of the roof in real time and control it. Compared with the traditional genetic algorithm, while ensuring the control effect, the calculation time is greatly reduced, the calculation efficiency is greatly improved, and as the optimization scale increases, the advantage of this algorithm in improving the calculation efficiency will become more obvious.

[0007] Technical solution:

[0008] A method for placing sensors and actuators for active noise control in helicopter cabins is proposed. Machine learning is introduced into the sensor / actuator optimization problem in active noise control in helicopter cabins. A machine learning-based intelligent optimization algorithm, BPNN-GA (Back Propagation Neural Network-Genetic Algorithm, BPNN-GA), is proposed. The algorithm mainly includes two processes: sensor optimization based on neural networks and actuator optimization based on genetic algorithms. Among them, machine learning is used to optimize the location and number of sensors, that is, a high-precision top plate vibration-radiation noise model is established using neural networks, and the main sound radiation surface elements are identified based on the model as the sensor layout points; while the genetic algorithm is used to optimize the actuators.

[0009] Specifically, the steps include:

[0010] Step 1: Divide the helicopter cabin top plate into a number of radiation surface elements based on the specifications of the actuator, and arrange acceleration sensors and MFC actuators at the centers of the radiation surface elements;

[0011] Step 2: Taking the helicopter cabin noise as the control object, a multi-line spectrum complex neural network model is established, the cabin wall panel vibration response obtained by the acceleration sensor and the cabin noise response obtained by the microphone are used as the input and output of the multi-line spectrum complex neural network model respectively, and the main sound radiation surface element is identified as the preferred arrangement point of the acceleration sensor;

[0012] Step 3: Based on the determination of the optimal layout points of the acceleration sensors, the layout points of the actuators are optimized using a genetic algorithm with the goal of reducing the noise level of the monitoring points, and finally the optimal layout points of the MFC actuators are obtained.

[0013] Preferably, step 2 comprises:

[0014] Step 2.1, establish a multi-line spectrum complex neural network model, and establish an independent sub-model for each line spectrum frequency of the wall panel vibration, and the sub-model retains the amplitude and phase information of the vibration noise response;

[0015] Step 2.2, modify the network weight coefficients based on the "back propagation" algorithm;

[0016] Step 2.3, repeat the gradient solution and weight coefficient update in step 2.2 until the preset maximum number of iterations is reached, and obtain the transfer matrix between the wall panel vibration and the radiation noise based on the relationship between the input and output of the neural network to obtain the radiation noise contribution of each surface element, and then identify the main sound radiation surface elements.

[0017] Preferably, in step 2.1, for a neural network including an input layer, H hidden layers and an output layer, the input vector of the BP neural network is X = [x1, x2, ..., x M ] T , which characterizes the vibration response of each radiation element of the top plate at ω = ω0, M is the total number of radiation elements; for the mth radiation element, its response is and are the real and imaginary parts of the response, respectively, including the amplitude and phase information of the vibration; the target vector of the BP neural network is Y = [y1, y2, ..., y N ] T , that is, the noise response of N monitoring points in the cabin at ω = ω0; similarly, for the nth monitoring point, its noise response is The relationship between the input and output of the neural network is obtained as follows:

[0018]

[0019] In the formula, is the actual output of the neural network, W h is the weight coefficient matrix of the hth layer, is the transfer matrix between panel vibration and radiated noise.

[0020] Preferably, in step 2.1, certain constraints are imposed on the bias matrix and the activation function: all bias items are set to 0, and the activation function adopts the Linear activation function.

[0021] Preferably, in step 2.2, the chain rule is used for "back propagation", and the mean square error is used as the loss function to calculate the gradient of the loss function for each parameter; based on the gradient descent method, the weight coefficients between the layers of the network are corrected along the negative gradient direction of the loss function, and the real and imaginary parts are simplified in combination with the imposed constraints, and a learning rate η is given to obtain the updated weight coefficient, which is expressed as:

[0022]

[0023] In the formula, for u j The conjugate of is the loss function E at u i The gradient at .

[0024] Preferably, in step 2.3, the transfer matrix between the wall panel vibration and the radiated noise is obtained based on the relationship between the input and output of the neural network in step 2.1. Find the radiation noise generated by M pixels at the monitoring point The radiation noise is converted to the complex plane for marking and superposition to obtain the sum of the vibration responses of each radiation surface element, that is, the total response vector; based on the total response vector, the contribution of each surface element to the overall radiation noise is calculated, and the comprehensive contribution of each surface element is further determined.

[0025] Preferably, in step 2.3, the comprehensive contribution of the mth bin is:

[0026]

[0027] In the formula, is the frequency ω k The radiation noise contribution of the mth bin at the nth monitoring point is,

[0028] is the phase weighting coefficient The sign function form is used to determine the contribution of radiation noise energy.

[0029] Preferably, in order to quantify the contribution of each surface element to the total radiation noise energy, the energy contribution rate of each radiation surface element is calculated based on the following formula:

[0030]

[0031] A threshold is set, and starting from the face element with the largest contribution, the energy contribution rate of each face element is accumulated in sequence until the cumulative contribution rate reaches or exceeds the set threshold. In this process, the face element whose energy contribution rate is accumulated to reach or exceed the set threshold is defined as the main sound radiation face element.

[0032] Preferably, in step 3, first, a set of actuator layout schemes are randomly generated as the initial population; then, based on the optimization objectives and constraints, a fitness function is set to quantitatively evaluate the performance of each layout scheme; finally, through selection, crossover and mutation operations, the individuals in the population are gradually iterated and optimized until the termination condition is met.

[0033] As a preferred method, a multi-line spectrum objective function is established to realize multi-objective optimization. Specifically, the position and number of actuators are optimized, that is, q actuators are used to replace the original p actuators, q<p, and the error signal of each measuring point is:

[0034] e m =d m +G mq u q

[0035] In the formula, u q is the active control quantity, that is, the driving signal of the q actuators after the optimization, which is a vector of q×1, G mq is the admittance matrix of the control channel;

[0036] By using the optimal control method, the optimal control quantity and the noise response of r monitoring points in the cabin are obtained, and the noise reduction effect after control is obtained as a single-line spectrum objective function, which is expressed as:

[0037]

[0038] In the formula, is the noise energy of the monitoring point without control, is the noise energy of the monitoring point after control,

[0039] p0 is the reference sound pressure;

[0040] The objective function of constructing multi-line spectrum is expressed as:

[0041]

[0042] In the formula, J i represents the noise reduction effect under the i-th line spectrum, w i is the weight of the objective function of the i-th line spectrum, and K is the number of total line spectra.

[0043] The beneficial effects of the present invention are:

[0044] (1) Compared with traditional genetic algorithms (TGA), the BPNN-GA algorithm (Back Propagation Neural Network-Genetic Algor ithm, BPNN-GA) proposed in the present invention significantly reduces the calculation time while ensuring the control effect, greatly improves the calculation efficiency, and as the optimization scale increases, the advantage of the algorithm in improving the calculation efficiency will become more obvious;

[0045] (2) Compared with the empirical selection, the BPNN-GA algorithm proposed in this invention focuses on the relationship between the sensor actuator position and the noise at the monitoring point, and has better noise control effects at the monitoring point and in the cabin. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below.

[0047] Figure 1 is an ASAC control strategy diagram of an embodiment of the present invention;

[0048] Figure 2 is a flow chart of a BPNN-GA algorithm according to an embodiment of the present invention;

[0049] Figure 3 This is a multi-line spectrum complex neural network model diagram of an embodiment of the present invention;

[0050] Figure 4 is a schematic diagram of a response vector according to an embodiment of the present invention;

[0051] Figure 5 is a diagram of an actuator optimization process based on a genetic algorithm according to an embodiment of the present invention;

[0052] Figure 6 is a diagram of an active control and noise monitoring system according to an embodiment of the present invention;

[0053] Figure 7 is a diagram of the arrangement of a sensor actuator according to an embodiment of the present invention;

[0054] Figure 8 1 is a plane noise level diagram of the ear position of an embodiment of the present invention, wherein (a) 360 Hz is uncontrolled, (b) 360 Hz is controlled, (c) 900 Hz is uncontrolled, and (d) 900 Hz is controlled. DETAILED DESCRIPTION

[0055] The embodiments of the present invention provide a method for making the purpose, technical solution and advantages of the present invention more clear, and the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0056] During the flight of a helicopter, the vibration of the main reducer will be transmitted to the cabin wall through the supporting structure, exciting the vibration of the wall panel and radiating noise into the cabin. This structural sound transmission is the main source of noise in the helicopter cabin. The wall panel is the main transmission path and direct radiation source of noise in the helicopter cabin. The top plate, as a key forced vibration area, is the main source of radiated noise. Reducing its vibration level can effectively reduce radiated noise and prevent vibration energy from being transmitted to other cabin wall panels. By integrating the actuator, sensor and helicopter top plate to form an intelligent active wall panel, the vibration of the top plate can be sensed and controlled in real time, thereby achieving the purpose of treating noise with vibration.

[0057] The active structural acoustic control (ASAC) of the present invention is as follows Figure 1 As shown in the figure, it includes radiation surface element division and sensor / actuator intelligent optimization algorithm. Among them, the top plate is directly excited by the main reducer through the supporting structure, radiating noise into the cabin. Based on the specifications and dimensions of the piezoelectric fiber composite (Macro Fiber Composite, MFC) actuator, the helicopter cabin top plate is divided into several radiation surface elements, an acceleration sensor is arranged at the center of the radiation surface element, and an MFC actuator is configured at the same position. At this time, several sensors and actuators on the top plate constitute the input and output of the active control system. Subsequently, the location and number of sensors are optimized by machine learning methods, modeling is carried out based on a data-driven approach, and a high-precision top plate vibration-radiation noise model is constructed using a neural network. Several microphones are arranged in the cabin according to the actual position of the passengers to simulate the sound pressure at the ears of the passengers in the cabin. Four concentrated load excitations are applied to the top of the cabin model to simulate the excitation load transmitted to the top plate by the reducer support / strut. Excitation is applied to the cabin model, and the acceleration value at the center of each radiation surface element is used to characterize the vibration response of the radiation surface element. The acceleration response data and noise response data are used as the input and output of the neural network respectively. The relationship between the wall panel vibration and the radiated noise is accurately fitted through multiple iterations. The radiation surface element that contributes most to the cabin noise, that is, the main sound radiation surface element, is identified using an intelligent optimization algorithm, and used as the sensor layout point. After the sensor layout plan is determined, the actuator is optimized based on a genetic algorithm with the goal of minimizing the noise at the monitoring point, aiming to ensure that the vibration of the control point is effectively suppressed while minimizing the cabin noise level.

[0058] Specifically, the present application discloses a method for deploying a sensor actuator for active noise control in a helicopter cabin, comprising three steps:

[0059] Step 1: Divide the helicopter cabin roof into several radiation surface elements based on the specifications of the actuator, and arrange the acceleration sensor and MFC actuator at the center of each radiation surface element. During the flight of the helicopter, the vibration of the main reducer causes the cabin wall panel to vibrate and radiate noise into the cabin. In order to reduce the noise level at the ears of the passengers, the acceleration sensor is used to obtain the vibration response of the cabin wall panel, and the microphone is used to obtain the cabin noise response, and the data is input into the active control and noise monitoring system.

[0060] Step 2, sensor optimization based on neural network. In order to reduce the control scale and ensure the noise reduction effect at the same time, it is necessary to identify and prioritize the main sound radiation surface elements that contribute more to the cabin noise. Taking the helicopter cabin noise as the control object, a multi-line spectrum complex neural network model is established. On this basis, the acceleration response data and noise response data are used as the input and output of the neural network respectively. Subsequently, the neural network model is trained, and the relationship between the hidden layer and the input and output is obtained based on the "forward propagation" process of the BP neural network. In order to optimize the network performance, the weight coefficient of the network is corrected based on the "backward propagation" algorithm. Further, based on the gradient descent method, the weight coefficients between each layer of the network are corrected in the negative gradient direction of the loss function. The above gradient solution and weight coefficient update are repeated until the preset maximum number of iterations is reached, and the relationship between the wall panel vibration and the radiation noise is accurately fitted. Further, based on the radiation noise generated by the M surface elements at the monitoring point, the vibration response of each radiation surface element is marked and superimposed on the complex plane to obtain the total response vector, the radiation noise contribution is calculated, and the main sound radiation surface element is identified as the sensor layout point.

[0061] Specifically,

[0062] Step 2.1 After the radiation surface elements are divided on the cabin roof, the noise inside the helicopter cabin is taken as the control object and the following is established: Figure 3 The multi-line spectrum complex neural network model shown in Figure 1. The characteristics of this model are: an independent model is established for each line spectrum frequency, while retaining the amplitude and phase information of the vibration noise response, thereby ensuring that the model has accurate physical characteristics.

[0063] from Figure 3 It can be seen that in the multi-line spectrum complex neural network model, the number of models is equal to the number of characteristic line spectra. For a neural network consisting of an input layer, H hidden layers and an output layer, the input vector of the BP neural network is X = [x1, x2, ..., x M ] T, which characterizes the vibration response of each radiation element of the top plate at ω = ω0, and M is the total number of radiation elements. For the mth radiation element, its response is and are the real and imaginary parts of the response, respectively, containing the amplitude and phase information of the vibration. The target vector of the BP neural network is Y = [y1, y2, ..., y N ] T , that is, the noise response of N monitoring points in the cabin at ω = ω0. Similarly, for the nth monitoring point, its noise response is

[0064] The relationship between the input and output of the neural network is obtained as follows:

[0065]

[0066] In the formula, is the actual output of the neural network, W h is the weight coefficient matrix of the hth layer, is the transfer matrix between panel vibration and radiated noise.

[0067] In order to meet the physical characteristics of acoustic vibration transmission and improve the accuracy and reliability of the model, certain constraints are imposed on the bias matrix and activation function. (1) All bias terms are set to 0. In a neural network, the bias term usually represents a fixed constant term that is independent of the input. However, in the actual acoustic vibration transmission process, there is no constant term that is independent of vibration and affects the propagation of radiated noise. If an arbitrary bias value is added only to make the neural network converge faster, it will cause the model to have a large error in prediction and violate the physical characteristics of acoustic vibration transmission. (2) The activation function uses a linear activation function. In a neural network, the activation function determines the output of the neuron, and its selection is usually based on the nature of the problem being solved. Since the present invention focuses on the linear transmission of acoustic vibration in subsequent simulations, nonlinear factors are not considered for the time being. Therefore, in order to ensure the accuracy of the model and prevent the distortion of the acoustic vibration transmission relationship, a linear activation function is selected.

[0068] Step 2.2, after the neural network "forward propagation", in order to optimize the network performance, the network weight coefficients are corrected based on the "backward propagation" algorithm. The premise of this correction process is to define a loss function, which measures the difference between the actual output of the network and the expected output. Here, the mean square error is selected as the loss function to quantify the degree of deviation between the predicted value and the true value, as follows:

[0069]

[0070] In the formula, the expected output y n The actual output of the network All are plural, * " indicates conjugate. Based on the gradient descent method, the weight coefficients between each layer of the network are corrected in the negative gradient direction of the loss function. Considering the influence of the real and imaginary parts, the gradient is expressed based on the chain rule. Combined with the imposed constraints, that is, the bias is set to 0 and the activation function is a linear activation function, the real and imaginary parts are simplified, and a learning rate η is given. The updated weight coefficients can be obtained as follows:

[0071]

[0072] In the formula, for u j The conjugate of is the loss function E at u i The gradient at .

[0073] Step 2.3, repeat the gradient solution and weight coefficient update of step 2.2 until the preset maximum number of iterations is reached, and obtain the transfer matrix between the panel vibration and the radiated noise based on formula (1): In order to obtain the contribution of radiation noise of each surface element, and then identify the main sound radiation surface elements.

[0074] From equation (1), we can deduce that the radiation noise generated by the mth surface element at the nth monitoring point is: It is expressed as:

[0075]

[0076] In the formula, is the transfer matrix The element in the nth row and the mth column represents the acoustic vibration transmission relationship between the vibration of the mth surface element and the noise of the nth monitoring point.

[0077] The comprehensive contribution of the mth panel is:

[0078]

[0079] In the formula, is the frequency ω k The radiation noise contribution of the mth surface element at the nth monitoring point is converted into energy through square processing. is the phase weighting coefficient The sign function form is used to judge the contribution of radiation noise energy. For positive time, A value of 1 indicates that the bin enhances the total noise energy and aggravates the overall noise level. When it is negative, A value of -1 indicates that the contribution of the face element to the total noise energy is negative, that is, it has a mutually canceling effect with the total noise energy.

[0080] Analogous to the use of cumulative contribution rate as the basis for judging the principal component in the field of machine learning, the optimization criteria for the main acoustic radiation surface elements of active noise control in the helicopter cabin are established. The specific implementation process is as follows: First, the comprehensive contribution of each radiation surface element is calculated based on formula (5). Secondly, in order to quantify the contribution of each surface element to the total radiated noise energy, the contribution rate of each radiation surface element is calculated based on formula (6):

[0081]

[0082] Furthermore, a certain threshold is set, usually in the range of 80% to 90%. Starting from the facet with the largest contribution, the energy contribution rate of each facet is accumulated in sequence until the cumulative contribution rate reaches or exceeds the set threshold. In this process, the facet whose energy contribution rate is accumulated to reach or exceed the set threshold is defined as the main sound radiation facet, and the sensor is arranged on it as a control point.

[0083] The selection of the threshold is a balancing process. A too high threshold may lead to an excessive number of selected main sound radiation elements, thereby increasing the control cost and control complexity. Therefore, in this embodiment, the threshold is initially set to 80%. At the same time, the scree plot analysis method in the field of machine learning is used to accurately adjust the threshold. A scree plot is drawn with the comprehensive contribution of radiation noise as the vertical coordinate, and the point where the contribution decreases in the figure begins to flatten out, that is, the "inflection point". The elements before these inflection points are usually the elements that contribute more to the total noise. By adjusting the threshold, it is ensured that the number of elements before the inflection point is not less than the number of elements determined by the cumulative energy contribution rate being greater than the threshold, thereby reducing the control cost while satisfying the control effect.

[0084] Step 3: Optimize actuators (including the number and location of actuators) based on genetic algorithms. Based on the determination of sensor locations, the genetic algorithm is used to optimize actuators with the goal of reducing the noise level at the monitoring point. Figure 5 As shown in the figure, first, a set of actuator layout schemes is randomly generated as the initial population. Then, based on the optimization objectives and constraints, a fitness function is set to quantitatively evaluate the performance of each layout scheme. Furthermore, through selection, crossover and mutation operations, the individuals in the population are gradually iterated and optimized until the termination condition is met. Finally, a set of optimized actuator layout schemes is obtained.

[0085] When using genetic algorithms for actuator optimization, the key lies in the construction of the objective function. In order to effectively reduce the noise level in the cabin, this paper establishes a multi-line spectrum objective function, aiming to achieve multi-objective optimization and ensure that the noise reduction effect under each line spectrum reaches a better level. When ω=ω0, the process of establishing the single-line spectrum objective function is as follows:

[0086] Assuming there are m sensors and p actuators, the error signal measured by the sensor can be expressed as:

[0087] e m =d m +G mp u p (7)

[0088] In the formula, e m is the error signal, i.e., the vibration response after the control of m sensors, which is an m×1 vector. m is the uncontrolled response, also an m×1 vector. p is the active control quantity, i.e. the driving signal of the actuator, which is a vector of p×1. mp is the admittance matrix of the control channel, which contains the one-to-one corresponding transfer relationship between m sensors and p actuators and is an m×p matrix.

[0089] At this time, according to the optimal control method in the frequency domain, the optimal control quantity can be obtained as:

[0090]

[0091] However, when there are m sensors and p actuators in the control system, m×p controller control channels are required. In engineering practice, too many actuators will lead to a large control scale, complicated control system, high control cost, and reduced control system stability. In order to reduce the control complexity and control cost of the system while ensuring control performance, this application considers optimizing the position and number of actuators (by optimizing the algorithm to reduce the number of actuators), that is, using q actuators to replace the original p actuators (q<p, and the preferred number of actuators is consistent with the number of sensors to ensure that the active control system is a completely deterministic system, thereby enhancing its stability), where P is an assumed value, which corresponds to the number of actuators in the control system before optimization in actual calculations. At this time, the error signal of each measuring point can be expressed as:

[0092] e m =d m +G mq u q (9)

[0093] In the formula, u q is the active control quantity, i.e., the driving signal of the q actuators after the optimization, which is a vector of q×1.mq is the admittance matrix of the control channel, which contains the one-to-one correspondence between m sensors and q actuators, and is an m×q matrix. By the optimal control method, the optimal control quantity and the noise response of the r monitoring points in the cabin can be obtained, and the noise reduction effect after control is obtained as the single-line spectrum objective function, as follows:

[0094]

[0095] In the formula, is the noise energy of the monitoring point without control, is the noise energy of the monitoring point after control, p0 is the reference sound pressure, and its size is 2×10 -5 Pa.

[0096] In order to comprehensively evaluate the noise reduction performance of the actuator layout scheme under multi-line spectra, a multi-line spectrum objective function is constructed based on formula (10), as follows:

[0097]

[0098] In the formula, J i represents the noise reduction effect under the i-th line spectrum, which can be calculated by formula (10). i is the weight of the objective function of the i-th line spectrum, which is used to adjust the relative importance of the noise reduction effects of different line spectra in the overall objective function, and its initial value is 1. K is the number of total line spectra.

[0099] In order to verify the feasibility and effectiveness of the BRNN-GA intelligent optimization algorithm proposed in the present invention, the inventors constructed an active control and noise monitoring system and conducted simulation analysis and verification. The top plate of the helicopter cabin model is divided into 63 sound radiation surface elements, and an acceleration sensor is arranged at the center of each radiation surface element to characterize the vibration response of the radiation surface element and serve as a feedback signal for active control. On this basis, an MFC actuator is configured at the same position. At this time, the 63 actuators and 63 sensors on the top plate constitute the input and output of the active control system. In order to monitor the noise level in the cabin, an in-cabin noise monitoring system is set up, and 8 monitoring points are arranged in the cabin, that is, 8 microphones are arranged to simulate the sound pressure at the ears of the passengers in the cabin. The active control and noise monitoring system is as follows Figure 6 Taking the established helicopter cabin model as the research object, the sensor actuator optimization is carried out based on the BPNN-GA algorithm, and the optimal solution is obtained, that is, the sensor actuator layout solution is as follows Figure 7As shown. The optimization results of the intelligent optimization algorithm were then simulated and verified. Table 1 shows the noise reduction simulation results of the helicopter cabin after sensor / actuator optimization. It can be seen that at 360Hz, the average sound pressure level of the monitoring point was reduced by 18.0dB. At 900Hz, the average sound pressure level of the monitoring point was reduced by 22.3dB. The cabin cavity noise level and cabin bulkhead vibration level were also suppressed. Figure 8 The simulation results of (a)-(d) show the plane noise level at the human ear position before and after the control is applied. It can be seen that after the active control is applied, the plane sound pressure level at the human ear position is significantly attenuated, especially in the middle area of ​​the plane, where the high noise is effectively suppressed, forming an obvious "quiet zone", achieving the purpose of reducing the noise level at the human ear of the driver and passengers. The sensor actuator scheme selected based on the BPNN-GA algorithm can achieve good control effects in the active control process, which not only effectively suppresses the noise level of each monitoring point, but also has a certain global vibration reduction effect and global noise reduction effect, thereby verifying the feasibility and effectiveness of the algorithm.

[0100] Table 1

[0101]

[0102]

[0103] The contents not described in detail in the specification of the present invention belong to the well-known technologies of professionals in the field.

[0104] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for deploying an actuator for active noise control sensor in a helicopter cabin, characterized in that: The steps include: Step 1: Divide the helicopter cabin top plate into a number of radiation surface elements based on the specifications of the actuator, and arrange acceleration sensors and MFC actuators at the centers of the radiation surface elements; Step 2, taking the helicopter cabin noise as the control object, establishing a multi-line spectrum complex neural network model, using the cabin wall panel vibration response obtained by the acceleration sensor and the cabin noise response obtained by the microphone as the input and output of the multi-line spectrum complex neural network model respectively, and identifying the main sound radiation surface element as the preferred arrangement point of the acceleration sensor; Step 3: Based on the determination of the optimal layout points of the acceleration sensors, the layout points of the actuators are optimized using a genetic algorithm with the goal of reducing the noise level of the monitoring points, and finally the optimal layout points of the MFC actuators are obtained.

2. The laying method according to claim 1, characterized in that: Step 2 includes: Step 2.1, establish a multi-line spectrum complex neural network model, and establish an independent sub-model for each line spectrum frequency of the wall panel vibration, and the sub-model retains the amplitude and phase information of the vibration noise response; Step 2.2, modify the network weight coefficients based on the "back propagation" algorithm; Step 2.3, repeat the gradient solution and weight coefficient update in step 2.2 until the preset maximum number of iterations is reached, and obtain the transfer matrix between the wall panel vibration and the radiation noise based on the relationship between the input and output of the neural network to obtain the radiation noise contribution of each surface element, and then identify the main sound radiation surface elements.

3. The laying method according to claim 2, characterized in that: In step 2.1, for a neural network consisting of an input layer, H hidden layers and an output layer, the input vector of the BP neural network is X = [x1, x2, ..., x M ] T , which characterizes the vibration response of each radiation surface element of the top plate at ω=ω0, M is the total number of radiation surface elements; For the mth radiation surface element, its response is and are the real and imaginary parts of the response, respectively, including the amplitude and phase information of the vibration; the target vector of the BP neural network is Y = [y1, y2, ..., y N ] T , that is, the noise response of N monitoring points in the cabin at ω = ω0; similarly, for the nth monitoring point, its noise response is The relationship between the input and output of the neural network is obtained as follows: In the formula, is the actual output of the neural network, W h is the weight coefficient matrix of the hth layer, is the transfer matrix between panel vibration and radiated noise.

4. The laying method according to claim 3, characterized in that: In step 2.1, certain constraints are imposed on the bias matrix and activation function: all bias items are set to 0, and the activation function uses the Linear activation function.

5. The laying method according to claim 4, characterized in that: In step 2.2, the chain rule is used for "back propagation", and the mean square error is used as the loss function to calculate the gradient of the loss function for each parameter. Based on the gradient descent method, the weight coefficients between the layers of the network are corrected along the negative gradient direction of the loss function, and the real and imaginary parts are simplified in combination with the imposed constraints. At the same time, a learning rate η is given to obtain the updated weight coefficient, which is expressed as: In the formula, for u j The conjugate of is the loss function E at u i The gradient at .

6. The laying method according to claim 2, characterized in that: In step 2.3, the transfer matrix between the panel vibration and the radiated noise is obtained based on the relationship between the input and output of the neural network in step 2.1 Find the radiation noise generated by M pixels at the monitoring point The radiation noise is converted to the complex plane for marking and superposition to obtain the sum of the vibration responses of each radiation surface element, that is, the total response vector; The contribution of each facet element to the overall radiation noise is calculated based on the total response vector, and the comprehensive contribution of each facet element is further determined.

7. The laying method according to claim 6, characterized in that: In step 2.3, the comprehensive contribution of the mth bin is: In the formula, is the frequency ω k The radiation noise contribution of the mth bin at the nth monitoring point is, is the phase weighting coefficient The sign function form is used to determine the contribution of radiation noise energy.

8. The laying method according to claim 7, characterized in that: In order to quantify the contribution of each surface element to the total radiated noise energy, the energy contribution rate of each radiating surface element is calculated based on the following formula: A threshold is set, and starting from the face element with the largest contribution, the energy contribution rate of each face element is accumulated in sequence until the cumulative contribution rate reaches or exceeds the set threshold. The face elements accumulated in this process are defined as the main sound radiation face elements.

9. The laying method according to any one of claims 1 to 8, characterized in that: In step 3, first, a set of actuator layout schemes are randomly generated as the initial population; then, based on the optimization objectives and constraints, a fitness function is set to quantitatively evaluate the performance of each layout scheme; finally, through selection, crossover and mutation operations, the individuals in the population are iteratively optimized step by step until the termination condition is met.

10. The laying method according to claim 9, characterized in that: A multi-line spectrum objective function is established to achieve multi-objective optimization. Specifically, the position and number of actuators are optimized, that is, q actuators are used to replace the original p actuators, q < p, and the error signal of each measuring point is: e m =d m +G mq u q In the formula, u q is the active control quantity, that is, the driving signal of the q actuators after the optimization, which is a vector of q×1. G mq is the admittance matrix of the control channel; By using the optimal control method, the optimal control quantity and the noise response of r monitoring points in the cabin are obtained, and the noise reduction effect after control is obtained as a single-line spectrum objective function, which is expressed as: In the formula, is the noise energy of the monitoring point without control, is the noise energy of the monitoring point after control, p0 is the reference sound pressure; The objective function of constructing multi-line spectrum is expressed as: In the formula, J i represents the noise reduction effect under the i-th line spectrum, w i is the weight of the objective function of the i-th line spectrum, and K is the number of total line spectra.

Images