A multi-source adaptive active noise control method and system for an automobile manufacturing workshop

Abstract

The application discloses a multi-source adaptive active noise control method for an automobile manufacturing workshop and a system thereof, relates to the technical field of noise control of an automobile manufacturing workshop, and comprises the following steps: acquiring a reference noise signal and an error noise signal; performing Fourier transform on the reference noise signal, and marking frequency components exceeding a preset threshold as target frequency components; extracting signal components according to the target frequency components and determining a current secondary channel transfer function; convolving the signal components with the current secondary channel transfer function and performing digital-to-analog conversion to obtain an analog driving signal; updating a coefficient according to a difference between the error noise signal and a preset target noise value; outputting the analog driving signal to an operating station through a loudspeaker; and taking the updated secondary channel transfer function as the current secondary channel transfer function of the next iteration. The application can reduce the amount of calculation, update the coefficient in a closed loop based on the actual residual noise at the operating station, weaken the influence of environmental interference, and keep noise reduction stable.

Description

Technical Field

[0001] This invention relates to the field of noise control technology in automobile manufacturing workshops. More specifically, this invention relates to a noise reduction method for automobile manufacturing workshops, and more particularly to a multi-source adaptive active noise control method and system for automobile manufacturing workshops. Background Technology

[0002] Automobile manufacturing workshops are equipped with process equipment such as stamping machines, welding robots, and conveyor lines. The stamping machine generates impact noise when it presses down during its stroke, the welding robot generates arc welding noise when it operates, and the conveyor line generates continuous mechanical noise during operation. The sound pressure level in the workshop is relatively high, and the noise exposure at the operating station directly affects the health of the workers. Therefore, workshop noise control has become an important part of the on-site environmental management of automobile manufacturing enterprises.

[0003] Currently, noise reduction in workshops mainly employs passive control measures such as sound insulation, sound absorption, and vibration reduction. Liang Changde, An Xueli, and Ou Yan, in their paper "Analysis of Noise and Vibration Pollution and Control Measures in a Stamping Workshop," disclosed test analyses of stamping press noise and vibration, calculated the noise's contribution to the factory boundary and sensitive points, determined the overall noise reduction amount, and proposed overall control schemes including sound insulation and vibration reduction. These passive measures have a good suppressive effect on mid-to-high frequency noise, but their effect on low-frequency components is limited. Furthermore, soundproof enclosures and sound-absorbing structures occupy a large space, making it difficult to cancel noise at the source. Regarding stamping process noise, Dai Yuande and Lu Xianfeng, in their paper "Stamping Process Noise Emission and Its Active Noise Reduction Methods," disclosed the mechanism of stamping process noise generated by sudden unloading and proposed active methods to reduce stamping process noise, such as progressive die two-step blanking, variable speed blanking, and die internal anti-ejection force blanking. These methods require modifications to the die structure and process parameters, and their versatility is limited by specific processes. Active noise control methods based on adaptive filtering achieve noise cancellation by generating secondary sound waves with similar amplitudes and opposite phases. However, they usually require multi-channel adaptive filtering of the full-band signal picked up by the reference microphone. The computational load increases with the processing bandwidth and the number of channels, limiting real-time performance. At the same time, background changes caused by airflow disturbances in the workshop and equipment start-up and shutdown can reduce the correlation between the reference signal and the error signal, causing the filter coefficient update to deviate from the actual acoustic path and resulting in fluctuations in noise reduction effect.

[0004] Therefore, it is necessary to provide a technical solution that can overcome the above-mentioned defects. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-source adaptive active noise control method and system for automobile manufacturing workshops, aiming to overcome the shortcomings of existing technologies such as large computational load of multi-channel adaptive filtering and inaccurate coefficient updates caused by workshop interference.

[0006] This invention aims to solve the problem of active noise control in the complex acoustic environment of automobile manufacturing workshops. Unlike the quiet environment of consumer electronics (such as noise-canceling headphones) or single noise sources, automobile manufacturing workshops are characterized by strong impact, multiple sound sources, time-varying nature, and high complexity. Traditional noise reduction technologies are completely ineffective in this scenario.

[0007] The technical problem to be solved by this invention is how to realize a real-time active noise reduction system with low latency, high stability, and high precision in harsh industrial environments characterized by high dynamics, multiple interferences, and strong impacts.

[0008] This invention is not only an improvement on the algorithm, but also a complete system engineering solution covering signal preprocessing, feature extraction, adaptive filtering, hardware layout, and system protection.

[0009] To achieve these objectives and other advantages of the present invention, according to one aspect of the present invention, a multi-source adaptive active noise control method and system for automobile manufacturing workshops are provided, comprising the following steps:

[0010] Step 1: Divide the automobile manufacturing workshop into multiple noise reduction zones. Place a set of reference microphones near the stamping machine, welding robot or conveyor line in each noise reduction zone, and place error microphones at the operating stations. Obtain the stamping cycle signal of the stamping machine. Based on the rising edge of the stamping cycle signal, extract the impact noise signal segment from the reference noise signal collected by the reference microphones. The time difference between the start time of the impact noise signal segment and the rising edge of the stamping cycle signal is less than a preset time threshold.

[0011] Step 2: Apply a Hanning window to the reference noise signal and perform a Fourier transform to obtain a frequency domain amplitude sequence. Mark the frequency components in the frequency domain amplitude sequence whose amplitudes are greater than a preset threshold as target frequency components; the preset threshold satisfies... , The threshold value is a preset value, and k is a preset multiple, where k is not less than 2 and not greater than 4. This is the average value of all amplitudes in the frequency domain amplitude sequence;

[0012] Step 3: Based on the frequency band where the target frequency component is located, select the corresponding current secondary channel transfer function from the preset transfer function library; the preset transfer function library is divided into a stamping machine transfer function group, a welding robot transfer function group, and a conveyor line transfer function group. The transfer functions of each group are measured at the operating station when the corresponding noise source is running alone; monitor the temperature value in the operating station, and when the change in temperature value is greater than the preset temperature threshold, perform temperature correction on the current secondary channel transfer function based on the physical model of sound speed changing with temperature;

[0013] Step 4: Filter the reference noise signal using a bandpass filter to obtain signal components, and convolve the signal components with the current secondary channel transfer function to obtain a secondary noise signal; perform digital-to-analog conversion on the secondary noise signal, and compress the peak amplitude of the secondary noise signal before conversion to ensure that it is not greater than the rated input amplitude of the speakers arranged around the operating station to obtain an analog drive signal;

[0014] Step 5: Distribute the analog drive signal to the loudspeaker array arranged around the operating station; according to the distance between each loudspeaker unit and the error microphone arranged at the operating station, perform amplitude distribution and delay compensation on the signal input to each loudspeaker unit so that the sound waves of each path are superimposed and canceled at the position of the error microphone.

[0015] Step 6: Based on the difference between the error noise signal collected by the error microphone and the preset target noise value, update the coefficients corresponding to the target frequency component in the current secondary channel transfer function to obtain the updated secondary channel transfer function, and use it as the current secondary channel transfer function for the next iteration.

[0016] Further, in step 1, the reference microphone group includes multiple reference microphone units uniformly arranged circumferentially along the noise source; a differential signal sequence is constructed by subtracting the signals collected by the multiple reference microphone units pairwise; the operating switching time of the noise source is identified based on the amplitude change of the differential signal sequence, so as to dynamically adjust the interception window of the impact noise signal segment.

[0017] Further, in step 2, the reference noise signal is divided into multiple signal segments. After performing a Fourier transform on each signal segment, the amplitudes of each signal segment at the same frequency are averaged to obtain an average frequency domain amplitude sequence. The number of target frequency components is counted. When the number of target frequency components is greater than a preset upper limit, the preset upper limit of frequency components are selected as target frequency components in descending order of amplitude. When the target frequency component of the current iteration is completely consistent with the target frequency component of the previous iteration and the number of consecutive iterations reaches the preset upper limit of the number of iterations, the preset multiplier k is reduced by a preset adjustment amount.

[0018] Furthermore, in step 4, multiple bandpass filters are used to filter the reference noise signal. Each bandpass filter corresponds to a target frequency component, and the bandwidth of each bandpass filter is no greater than 1 / 2 of the frequency interval between two adjacent target frequency components. When the frequency interval between two adjacent target frequency components is less than a preset interval threshold, the two adjacent target frequency components are merged into one target frequency component, and the frequency band range of the bandpass filter is redefined based on the merged target frequency component.

[0019] Furthermore, in step 3, when the target frequency component is simultaneously located in the frequency bands corresponding to multiple transfer function groups, the current secondary channel transfer function H is obtained by summing using the following formula:

[0020]

[0021] Let J be the secondary channel transfer function of the corresponding frequency band in the j-th transfer function group, where J is the number of transfer function groups.

[0022] Further, in step 4, a piecewise convolution operation is performed on the signal components and the current secondary channel transfer function. The length of each segment of the piecewise convolution operation is the same as the window length of the Hanning window, and the interval between the starting positions of two adjacent segments is 1 / 2 of the window length. The outputs of the piecewise convolution operation are combined in an overlapping addition manner. When overlapping addition, the outputs of two adjacent segments are averaged in the overlapping interval and then added to obtain the secondary noise signal.

[0023] Further, in step 6, the update difference is calculated. Where T is the preset target noise value, and RMS is the root mean square value of the error noise signal within a preset time period; when |e| is greater than the preset update threshold, the coefficient is updated according to the preset step size μ, where the preset step size μ = μ0 × |e|, and μ0 is the preset base step size; when the number of times |e| is greater than the preset update threshold reaches the preset upper limit, the coefficient corresponding to the target frequency component in the current secondary channel transfer function is multiplied by the preset attenuation coefficient α ∈ [0.5, 0.9];

[0024] Further, in step 5, the loudspeaker array consists of multiple loudspeaker units, which are evenly arranged around the center of the operating station, with the same angle between any two adjacent loudspeaker units and the center of the operating station; the delay of the i-th loudspeaker unit... satisfy:

[0025]

[0026] in, Let be the distance between the i-th loudspeaker unit and the error microphone, and c be the preset sound velocity.

[0027] Furthermore, in step 4, when the amplitude of the secondary noise signal is not greater than the preset lower limit of amplitude within a preset time period, the analog drive signal is set to a zero amplitude signal, wherein the preset lower limit of amplitude is 1% of the rated input amplitude of the speaker; zero-fill interpolation is performed on the analog drive signal, and the interpolation multiple is set to 2 to improve the smoothness of the output signal;

[0028] According to one aspect of the present invention, the present invention also provides a multi-source adaptive active noise control system for automobile manufacturing workshops, comprising:

[0029] A reference microphone array is placed near the noise source in the workshop to collect reference noise signals;

[0030] A beat sensor, connected to the stamping press, is used to acquire the stamping beat signal;

[0031] An error microphone is placed at the operating station to collect error noise signals.

[0032] A loudspeaker array, arranged around the workstation, is used to output noise immunity;

[0033] A signal processing device is connected to the reference microphone group, the beat sensor, the error microphone, and the speaker array, respectively, and the signal processing device is configured to execute the multi-source adaptive active noise control method for automobile manufacturing workshops described above.

[0034] The present invention has at least the following beneficial effects:

[0035] This invention performs bandpass extraction, convolution operations, and coefficient updates only on frequency components with amplitudes greater than a preset threshold. The processing range is concentrated on the main noise frequency bands, thus reducing the computational load and meeting the real-time processing requirements of active noise reduction in the workshop. Coefficient updates are based on the difference between the error noise signal and the preset target noise value. The coefficient adjustment direction always aims to reduce the actual residual noise at the operating station. Background changes caused by airflow disturbances and equipment start-ups / stops within the workshop are incorporated into the difference feedback, making it less likely for the coefficient update process to deviate from the actual acoustic path, thereby avoiding fluctuations in noise reduction effects due to inaccurate updates. The updated secondary channel transfer function serves as the current secondary channel transfer function for the next iteration, continuously adjusting with changes in stamping cycle time and equipment start-ups / stops, ensuring stability during production cycle switching. Furthermore, it is designed for workshop noise sources such as stamping machines, welding robots, and conveyor lines, and is tailored to the actual operating scenarios of automotive manufacturing workshops.

[0036] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0037] Figure 1 This is a flowchart of a multi-source adaptive active noise control method for automobile manufacturing workshops according to this application;

[0038] Figure 2 This is a framework diagram of a multi-source adaptive active noise control system for automobile manufacturing workshops, as described in this application. Detailed Implementation

[0039] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.

[0040] It should be understood that terms such as "having," "comprising," and "including" used in the embodiments of this application do not exclude the presence or addition of one or more other elements or combinations thereof. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of components in a specific posture. If the specific posture changes, the directional indication will also change accordingly. When an element is referred to as "fixed to" or "set on" another element, it can be directly on the other element or may have an intervening element present. When an element is referred to as "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element through an intervening element. Descriptions involving "first," "second," etc., in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features.

[0041] It should be noted that the technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0042] Embodiments of this application provide a multi-source adaptive active noise control method for automobile manufacturing workshops, comprising the following steps:

[0043] Step 1: Sound Source Monitoring and Signal Acquisition: Acquire the reference noise signal collected by the reference microphone and the error noise signal collected by the error microphone. The reference microphone is placed near the noise sources in the automobile manufacturing workshop, such as stamping machines, welding robots, and conveyor lines. The error microphone is placed at the operating station. The reference microphone is used to pick up sound pressure changes before the equipment operating noise reaches the work area. Its diaphragm converts sound pressure into voltage. The reference microphone is placed close to the noise source to ensure that the picked-up waveform contains the main vibration components of the noise source and that the propagation path loss is small. The error microphone is used to pick up the actual remaining sound energy at the operating station after noise reduction processing. Its output serves as the basis for evaluating the noise reduction results and correcting control parameters. Both the reference microphone and the error microphone are connected to the acquisition channel of the noise reduction control unit through shielded cables. The acquisition channel is synchronously triggered according to a preset sampling frequency to ensure that the two signals are aligned in time. The preset sampling frequency is, for example, 16000 Hz or 32000 Hz. The noise reduction control unit is an industrial control device installed in the workshop, used to perform signal processing and control calculations;

[0044] Step 2: Spectral Feature Extraction: Apply a Hanning window to the reference noise signal and perform a Fourier transform to obtain a frequency domain amplitude sequence. Mark the frequency components with amplitudes greater than a preset threshold in the frequency domain amplitude sequence as target frequency components. The Hanning window is used to suppress spectral leakage caused by truncation. The duration of the Hanning window is matched with the preset sampling frequency, for example, corresponding to 256 or 512 sampling points. The Fourier transform decomposes the time domain waveform into the amplitude and phase of each frequency component. The preset threshold is pre-calibrated according to the energy level of the main noise frequency band in the workshop to distinguish between the main noise components and the background components. The target frequency component corresponds to the main noise frequency band with concentrated energy in the workshop, such as the low frequency band of stamping impact noise or the mid frequency band of conveyor drive noise.

[0045] Step 3: Multi-source transfer function matching and correction: Based on the target frequency components, determine the current secondary channel transfer function. The current secondary channel transfer function is selected from the preset transfer function library or is the updated secondary channel transfer function. The preset transfer function library stores the secondary channel transfer functions for each frequency between the noise source and the operating station. The secondary channel transfer function describes the acoustic path characteristics of the analog drive signal propagating through the loudspeaker and air to the error microphone position. The transfer function library is measured and stored point by point according to frequency during the workshop debugging phase. The selection operation is performed once in the current iteration.

[0046] Step 4: Noise Generation and Amplitude Limitation: The bandwidth of the bandpass filter is determined based on the target frequency component. The passband of the bandpass filter extends to both sides with the target frequency component as a reference, used to extract the signal component corresponding to the target frequency component from the reference noise signal, thereby filtering out background components outside the bandwidth. The signal component is convolved with the current secondary channel transfer function to obtain the secondary noise signal. The convolution operation weights and superimposes the signal components according to the impulse response of the acoustic path, used to simulate the amplitude and phase state of the secondary sound wave reaching the error microphone position. The input of the convolution operation is the signal component and the current secondary channel transfer function, and the output is the secondary noise signal. Subsequently, the secondary noise signal is converted from digital to analog, and the peak amplitude of the secondary noise signal is compressed before conversion to ensure that it is not greater than the rated input amplitude of the loudspeakers arranged around the operating position, thus obtaining the analog drive signal. The sampling rate of the digital-to-analog conversion is matched with the aforementioned preset sampling frequency to ensure the waveform accuracy of the analog drive signal.

[0047] Step 5: Spatial Sound Field Reconstruction Output: The analog drive signal is output to the operating station through speakers arranged around it. The speakers convert the analog drive signal into secondary sound waves. The secondary sound waves spatially superimpose with the sound waves directly reaching the noise source at the operating station. After superposition, the total sound pressure decreases compared to before superposition. The updated secondary channel transfer function becomes the current secondary channel transfer function for the next iteration. The next iteration starts by reacquiring the reference noise signal and the error noise signal, thus continuously updating the secondary channel transfer function according to the changes in workshop conditions. Taking the working scenario of a stamping machine as an example, impact noise is generated at the moment the stamping machine presses down. The reference microphone picks up the impact waveform before the noise reaches the operating station. The noise reduction control unit generates the corresponding analog drive signal based on this. The secondary sound waves cancel each other out at the same time as the impact sound waves reach the operating station, thereby shortening the time that the operating station is subjected to high sound pressure.

[0048] Step 6: Adaptive Coefficient Update: Based on the difference between the error noise signal and the preset target noise value, update the coefficients in the current secondary channel transfer function corresponding to the target frequency component to obtain the updated secondary channel transfer function. The preset target noise value is the residual noise level that is expected to be maintained at the operating station. The difference is used to indicate the degree of deviation between the current cancellation effect and the expected level. The coefficients are adjusted in the direction of decreasing difference to gradually approach the optimal cancellation state. The updated coefficients are immediately used for the convolution operation in subsequent iterations.

[0049] In existing technologies, workshop noise reduction typically performs multi-channel adaptive filtering on the full-band noise signal picked up by the reference microphone. The processing range covers all frequency components, and the computational load increases with the bandwidth and the number of channels. Simultaneously, background changes caused by airflow disturbances and equipment start-ups and shutdowns within the workshop reduce the correlation between the reference signal and the error signal, causing the filter coefficient updates to deviate from the actual acoustic path. Furthermore, once the filter parameters are fixed during the debugging phase, they are no longer adjusted according to production cycle changes. This embodiment performs extraction, convolution, and coefficient updates only on frequency components whose amplitude exceeds a preset threshold. The processing range is concentrated on the main noise frequency band, thus reducing the computational load. Simultaneously, the coefficient updates are driven by the difference between the residual noise picked up by the error microphone and the preset target noise value. Coefficient adjustments are always based on the actual residual noise at the operating station, thereby maintaining accurate tracking of the cancellation path even when workshop operating conditions fluctuate.

[0050] Furthermore, in step 1, the automobile manufacturing workshop is divided into multiple noise reduction zones. These zones are defined according to the floor plan layout of the equipment within the workshop. Each zone covers one or a group of adjacent devices. The boundaries of the zones are set in the soundproof areas between the devices. A set of reference microphones is arranged within each noise reduction zone. The distance between the reference microphones and the noise source is no greater than a first preset distance threshold, such as 0.5 m or 1 m. This ensures that each noise reduction zone independently picks up the noise components of its corresponding device, and that the reference microphones in adjacent zones do not overlap. The reference microphone signals within each zone are sent to the same acquisition bus after gain calibration. Gain calibration is used to eliminate sensitivity differences between channels. The stamping cycle signal of the stamping press is acquired. This signal is output by the stamping press controller and input to the noise reduction control unit in the form of level transitions to indicate the timing of the stamping stroke. The control unit counts the rising edges and records the time interval between adjacent rising edges. Based on the rising edge of the stamping cycle signal, an impact noise signal segment is extracted from the reference noise signal. The extraction operation uses the reference noise signal as input and the impact noise signal segment as output. The starting position of the extraction is determined by shifting the time corresponding to the rising edge forward by a preset time threshold. The time difference between the starting time of the impact noise signal segment and the rising edge of the stamping cycle signal is less than the preset time threshold, such as 5 ms or 10 ms. This distinguishes the impact noise generated by stamping from the continuous noise generated by the conveyor line. The length of the impact noise signal segment meets the following requirements: The time interval between two adjacent pressing strokes should not exceed half of the total time interval to avoid overlapping of adjacent impact segments. The ending position of the cut-off segment is determined by extending the length of the impact segment from the starting position, and the length of the impact segment is calculated by the time interval between adjacent rising edges. An error microphone is placed in the operating station. The operator's ear position is determined by the height of the workstation seat and the operating posture. The error microphone is mounted on a bracket near this position. The distance between the error microphone and the operator's ear position is not greater than a second preset distance threshold, such as 5cm or 10cm, to ensure that the sound pressure picked up by the error microphone is close to the sound pressure actually felt by the operator. The reference noise signal includes an impact noise signal segment and a continuous noise signal segment. The continuous noise signal segment is generated by the conveyor line. Steps 2 to 6 are performed on the impact noise signal segment and the continuous noise signal segment, respectively. That is, the impact component and the continuous component independently complete frequency domain labeling, bandpass extraction, convolution generation, and coefficient updating. The two signal paths share the same set of bandpass filters, transfer function library, and loudspeakers to avoid mutual interference between the two types of noise characteristics. The convolution and coefficient updates of the two signal paths are performed in a time-division manner within the control unit, and the outputs are merged into a unified secondary noise signal in chronological order. When the time interval between adjacent rising edges of the stamping cycle signal changes, the impact noise signal segment is re-triggered according to the changed time interval, and the re-triggered impact noise signal segment is used as the input of the Fourier transform in step 2. The change in the stamping cycle corresponds to the adjustment of the stamping frequency or the switching of the production line cycle. The re-triggering ensures that the length and phase of the impact segment always match the current stamping rhythm, thereby maintaining the tracking and cancellation of the impact noise. During the re-triggering, the temporary data of the old impact segment is cleared, and the input of the Fourier transform is updated with the new trimming result. Taking a station where the stamping machine and the conveyor line run in parallel as an example, the stamping machine generates an impact segment each time it presses down, and the conveyor line continuously generates a continuous segment. After processing by both separately, they are combined to form a complete noise reduction output.

[0051] In this embodiment, the impact noise signal segment is extracted based on the rising edge of the stamping cycle signal. The start time of the impact segment directly corresponds to the timing of the stamping action. When the cycle changes, the segment is re-extracted according to the new interval. The distance between the error microphone and the ear position is constrained. The cycle information comes from the stamping machine controller and can be obtained without adding additional sensors. Thus, the corresponding secondary sound wave can still be generated in advance when the stamping frequency changes, and the timing of impact noise cancellation at the operating station is more accurate.

[0052] It is important to note that this invention addresses the challenge of synchronizing and tracking "strong non-stationary impact noise." The noise generated by a stamping press is a transient impact signal (pulse-like), characterized by short duration (milliseconds), high energy, and a wide frequency spectrum. Traditional noise reduction systems are typically designed for steady-state noise (such as engine hum), and their convergence speed cannot keep up with the changes in impact noise. This results in either insufficient reaction time or "overshoot" oscillations after the impact, and the inability to anticipate when the noise occurs, making synchronization difficult. Least mean square algorithms require multiple iterations to adapt after the impact, by which time the impact may have already ended, leading to slow convergence. This invention utilizes the stamping press's cycle signal (rising edge) as a "predictive" trigger signal to precisely capture the impact noise segment, achieving "pre-aiming" noise reduction.

[0053] This embodiment achieves "pre-aiming" impact cancellation and millisecond-level transient convergence, breaking through the physical limits of non-stationary noise control. In the field of active noise control, for transient impact noise generated by a stamping press, the convergence speed of the conventional FxLMS algorithm usually lags behind the noise change, resulting in the system only being able to perform "post-event compensation," and producing a ringing effect lasting hundreds of milliseconds after the impact. It is generally believed by those skilled in the art that, given the unavoidable physical channel delay, it is impossible to completely eliminate the "pre-response" of impact noise. This invention introduces the rising edge of the stamping cycle signal as a hardware trigger source, utilizing the physical characteristic that the mechanical electrical signal (millisecond level) is much faster than the acoustic propagation speed, to achieve "pre-aiming" of the impact noise. The system can generate an inverse signal before the sound wave reaches the operating position, achieving near "zero-delay" cancellation; more importantly, combined with the variable step size update and attenuation mechanism of this invention, the system can converge instantaneously after the impact peak, completely eliminating the unpleasant oscillation aftereffects in traditional algorithms.

[0054] In one embodiment, in step 1, each set of reference microphones includes multiple reference microphone units. These units are installed at equal intervals along the circumference of the corresponding noise source, avoiding the sweeping area of ​​moving parts of the equipment. The sensitivity of each reference microphone unit is uniformly calibrated, and the differential results of each unit after calibration are unaffected by sensitivity differences. The multiple reference microphone units are arranged circumferentially around the corresponding noise source, with the same angle between adjacent reference microphone units and the noise source. The number of reference microphone units is not less than 4 and not more than 8. This uniform circumferential arrangement is used to pick up sound waves radiated from the noise source from different directions, thereby providing multi-angle information for determining the operating status of the noise source. The differential signal sequence is obtained by subtracting the signals collected by the multiple reference microphone units pairwise, and the differential value satisfies... ,Right now Let be the difference between the signal acquired by the i-th reference microphone unit and the signal acquired by the j-th reference microphone unit. The signal acquired by the i-th reference microphone unit. For the signal acquired by the j-th reference microphone unit, differential analysis is performed on the same time base. After time alignment, the signals from each unit are subtracted pairwise. The differential operation is used to eliminate the environmental background components common to all units, retain the differential components related to the changes in the location of the noise source, thereby amplifying the characteristics of the noise source state changes. The combination of pairwise subtractions covers all unit pairs, and the differential sequence is arranged in chronological order, with the number of differential values ​​corresponding to the number of unit pairs. When the amplitude change of the differential signal sequence exceeds a preset threshold, the moment with the largest amplitude change in the differential signal sequence is taken as the noise source's switching moment. The preset threshold is, for example, 0.2 or 0.3 times the peak amplitude of the differential signal. The amplitude change is measured by the local peak of the differential signal, and the moment corresponding to the maximum peak is the switching moment. The switching moment is used to determine the start and end times of the impact noise signal segment. The time interval between the start and end times of the impact noise signal segment is no greater than a preset duration threshold, for example, 0.5 s or 1 s. When the noise source starts or stops, the sound field structure picked up by each unit changes, and the differential amplitude peaks at the moment of switching. The determined start and end times correspond to the actual start and stop times of the equipment. The determined start and end times are sent back to the interception operation. The interception operation re-divides the impact noise signal segment according to the new start and end times, thus independently determining the boundary of the impact segment without relying on the equipment control signal. After the switching moment is determined, the start and end times of the impact segment are updated accordingly, and the updated results take effect in the next iteration. When the amplitude change of the differential signal sequence is no greater than a preset change threshold, the start and end times of the impulse noise signal segment remain unchanged, and the impulse noise signal segment from the previous iteration is used as the impulse noise signal segment for the current iteration. If the sound field does not change significantly, the truncation result from the previous iteration is used, thus avoiding repeated adjustments to the truncation boundary due to small fluctuations. The time interval between the impulse noise signal segment and the continuous noise signal segment is used as the noise source silence duration. When the silence duration is no greater than a preset silence threshold, the continuous noise signal segment is merged into the impulse noise signal segment. The preset silence threshold is, for example, 0.2 s or 0.3 s. A silence time that is too short indicates that the impulse noise and continuous noise are difficult to distinguish in time. After merging, the length of the impulse noise signal segment is correspondingly extended. The merging result serves as a unified input for subsequent steps, thus avoiding discontinuities in processing at the boundary. After merging, the continuous noise signal segment no longer undergoes separate subsequent steps; the merged segment is processed in the same way as the impulse noise signal segment. Taking the press start-up scenario as an example, the sound field difference picked up by each reference microphone increases at the moment the press starts, and the amplitude of the differential signal reaches a peak at the start-up moment. Based on this, the control unit determines the start time of the impact noise signal segment, so that the corresponding secondary sound wave is prepared before the press has completed its first press.

[0055] In this embodiment, the signals picked up by multiple reference microphones are differentially divided pairwise. The moment corresponding to the peak value of the differential amplitude is the switching moment of the noise source. The start and end times of the impact segment are determined independently based on this, no longer relying on the equipment control signal. When the silence time is too short, the continuous segment is incorporated into the overall processing of the impact segment. The processing at the boundary is smoother. The differential method uses the existing microphone signal without adding hardware. The judgment of the switching moment is based on the change of the sound field itself, which is suitable for various equipment. The impact noise at the moment of equipment start-up and shutdown can also be included in the interception range in a timely manner.

[0056] Presses and welding robots are not constantly operating; they are in a state of frequent start-stop and switching. The system needs to sense these states to adjust the noise reduction strategy. When the equipment is idle, the system is still running at full load, wasting energy. The impact at the moment of equipment start-up is easily missed or misjudged. This application uses the differential signals of multiple reference microphones to accurately capture the switching moment of noise sources, achieving "motion-based" noise reduction.

[0057] When stamping, welding, and conveyor lines operate alternately, noise characteristics change drastically. Traditional systems often fail when parameters are adjusted under one operating condition and switched to another, requiring technicians to manually switch modes or design extremely complex fuzzy logic controllers. This invention utilizes differential signal sequences to accurately capture the switching moments of noise sources. Combined with the classification and processing of impact and continuous noise, it achieves automatic decoupling and rapid switching of characteristic parameters of multiple noise sources. The system can automatically match the corresponding transfer function set at the moment of start-up and shutdown of different equipment. This "follow-up" noise reduction allows a single system to seamlessly cover all operating conditions in the workshop without manual intervention. Its adaptive switching smoothness and accuracy far exceed traditional solutions based on simple threshold judgments, resulting in unexpected effects.

[0058] In one embodiment, in step 2, the preset threshold satisfies That is, the preset threshold is the product of a preset multiple and the average amplitude, where the preset multiple k is not less than 2 and not greater than 4, and the average amplitude... The average amplitude is the sum of all amplitudes in the frequency domain amplitude sequence, obtained by summing the amplitudes of all frequency points in the frequency domain amplitude sequence and dividing by the number of frequency points. It reflects the overall level of current noise energy. The preset multiple is calibrated during workshop debugging, based on the amplitude difference between the background noise and the main noise frequency band. The preset threshold adjusts adaptively accordingly to avoid the fixed threshold failing when the total noise energy changes. The reference noise signal is divided into multiple signal segments, starting from the beginning of the reference noise signal and proceeding sequentially according to the window length. Adjacent segments do not overlap, and the number of segments covers the preset sampling duration. The length of each signal segment is the same as the Hanning window length, for example, corresponding to 256 or 512 sampling points. A Fourier transform is performed on each signal segment independently. The transform results are temporarily stored in segment order. Each signal segment is multiplied by the Hanning window before the Fourier transform is performed. The frequency domain amplitude sequence of each signal segment satisfies... ,in Let f be the amplitude of the frequency domain amplitude sequence at frequency f. For the nth sample value of the reference noise signal, Let f be the window coefficient of the Hanning window at the nth sampling point, N be the number of sampling points corresponding to the window length of the Hanning window, f be the frequency index, and j be the imaginary unit. The Fourier transform converts each segment from the time domain to the frequency domain, thus obtaining the energy distribution of each segment at different frequencies. The amplitudes of the frequency domain amplitude sequences of each signal segment at the same frequency are averaged. This averaging is performed at the same frequency index; the amplitudes of each segment at that frequency index are added together and divided by the number of segments to obtain the average frequency domain amplitude sequence. This average is used to smooth the amplitude fluctuations caused by noise between segments, thus obtaining a more stable spectrum estimate. When the number of segments is large, the average result tends to a stable spectral profile. Frequency components with amplitudes greater than a preset threshold in the average frequency domain amplitude sequence are marked as target frequency components. When the number of target frequency components is not greater than a preset upper limit, all frequency components with amplitudes greater than the preset threshold are marked as target frequency components. The preset upper limit is, for example, 16 or 32. When the number of target frequency components is greater than the preset upper limit, the preset upper limit of frequency components are selected as target frequency components in descending order of amplitude. The upper limit is used to constrain the size of the target frequency components, thereby controlling the number of frequency bands for convolution operations and coefficient updates within a preset range, thus prioritizing the processing of components with higher energy when there are many noise spectrum components. When the target frequency component of the current iteration is completely consistent with the target frequency component of the previous iteration, steps 3 to 5 are skipped, and step 6 is executed directly. That is, when the target frequency component has not changed, bandpass extraction, transfer function selection, and convolution operations are not repeated. Step 6 directly uses the current secondary channel transfer function of the previous iteration and the latest error noise signal as input, thereby avoiding redundant calculations. During the skip period, the signal component corresponding to the target frequency component and the secondary channel transfer function remain at the value of the previous iteration. When the target frequency component of the current iteration is inconsistent with the target frequency component of the previous iteration, steps 3 to 5 are executed to redetermine the signal component and the secondary channel transfer function corresponding to the new target frequency component. When the number of consecutive iterations with completely consistent target frequency components reaches the preset iteration limit, the preset multiplier is reduced by a preset adjustment amount. The preset iteration limit is, for example, 50 or 100, and the preset adjustment amount is, for example, 0.2 or 0.5. After the preset multiplier is reduced, the preset threshold decreases accordingly. Frequency components that were previously below the threshold re-enter the candidate range, and the target frequency component set is updated accordingly, thereby allowing more frequency components to enter the target frequency component set and preventing long-unchanged frequency bands from masking newly emerging noise components. The relabeled target frequency components participate in steps 3 to 5 in the next iteration. Taking the intermittent operation of a welding robot as an example, the robot generates high-frequency noise during welding. When the robot stops, the spectral components decrease, and the target frequency components change accordingly, triggering the re-execution of steps 3 to 5, thereby maintaining the tracking of noise spectrum changes during the robot's start-up and shutdown process;

[0059] This embodiment generates a preset threshold based on the average amplitude as a multiple. The threshold adaptively increases or decreases with the noise energy level. When the target frequency component remains unchanged, it skips repeated extraction and convolution steps, ensuring the processing range remains consistent with the actual noise frequency band, thus reducing computational overhead. Frequency bands that have remained unchanged for a long time are reconsidered by reducing the preset multiple. After the threshold becomes adaptive, manual periodic calibration is unnecessary, thus balancing targeted processing with following spectral changes. In a workshop, there are multiple noises simultaneously, such as those from stamping machines (impact), welding robots (high-frequency intermittent), and conveyor lines (low-frequency continuous). These noises overlap, and their frequency components may overlap. If noise reduction is applied uniformly to all frequencies, the entire system may become unpredictable due to unstable control of a single noise source, leading to mutual interference. Furthermore, calculating frequencies where no noise exists wastes computing power and resources. This application identifies the target frequency component using FFT and processes only the main noise frequencies. By establishing independent transfer function libraries for stamping, welding, and conveyor lines, different acoustic models are called for different types of noise sources.

[0060] In one embodiment, in step 4, the bandpass filter consists of multiple sets of bandpass filters, each set corresponding to a target frequency component. That is, each target frequency component is equipped with an independent filtering channel. The bandpass filters operate in parallel, filtering different frequency bands of the reference noise signal simultaneously. The filtering results are output to their respective temporary storage areas. Each set of filtering operations is independent, and the control unit schedules them by set. The sets do not share temporary storage areas. The bandwidth of each set of bandpass filters is no greater than half the frequency interval between two adjacent target frequency components. The passbands of adjacent sets of bandpass filters do not overlap. The bandwidth is determined by the frequency interval between adjacent target frequency components; the smaller the interval, the smaller the bandwidth. The lower limit of the bandwidth is determined by the filter order; with a fixed order, the bandwidth is narrower and the transition band is steeper. The constraints of bandwidth and interval ensure that the passbands of each set are staggered on the frequency axis, thereby avoiding duplicate extraction caused by the same frequency component entering two adjacent sets of filters simultaneously. The reference noise signal is input into each set of bandpass filters, and the bandpass filtering of each set of bandpass filters satisfies... ,in The signal component at time t, Let m be the filter coefficient of the bandpass filter. Using the sampled value of the reference noise signal at time tm as a reference, and M as the order of the bandpass filter, the filtering process performs a point-by-point weighted summation of the reference noise signal according to the filter coefficients to extract the signal components near the target frequency component. When the frequency interval between two adjacent target frequency components is less than a preset interval threshold, the two adjacent target frequency components are merged into one target frequency component. Merging is performed before bandwidth determination. The number of target frequency components after merging is reduced, and the number of bandpass filter groups is reduced accordingly. The preset interval threshold is, for example, 10 Hz or 20 Hz. The frequency range of the merged target frequency component covers the frequency range of the two target frequency components before merging. The frequency band range of the bandpass filter is redefined based on the merged target frequency component. Merging is used to avoid the passbands of the two filter groups squeezing each other when the frequency interval is too small, thereby covering the adjacent frequency band with a wider single filter group. The redefined bandwidth is no greater than half the frequency interval between the merged target frequency component and the adjacent target frequency component. Multiple signal components are arranged in ascending order of frequency according to the target frequency component and used as input for the convolution operation in step 4. The arrangement order corresponds to the subsequent convolution operation order, thus ensuring that the secondary noise signals of each frequency band are generated in an orderly manner according to frequency. The arranged signal components are written into the input queue of the convolution operation in sequence, and the order of the input queue is consistent with the arrangement order. Each signal component carries a group identifier that is the same as the corresponding target frequency component. The group identifier is consistent with the number of the target frequency component. Each data block of the signal component carries this identifier. The group identifier is used to determine the coefficients corresponding to the target frequency component in step 6. That is, when updating the coefficients, the corresponding frequency band and the corresponding transfer function coefficients are located based on the group identifier, thereby avoiding confusion of coefficients in different frequency bands. The coefficient table uses the group identifier as the row index. The bandpass filter is a finite impulse response filter. The order of the finite impulse response filter is not greater than a preset upper limit of order, such as 64 or 128. The finite impulse response filter has linear phase characteristics and a fixed output delay, thereby ensuring that the signal components do not produce phase distortion during the filtering process, which is convenient for matching with the phase characteristics of the secondary channel transfer function. Taking a stamping station with multiple frequency bands of noise as an example, the stamping impact frequency band and the conveyor transmission frequency band are extracted by two sets of bandpass filters respectively. When the distance between the two frequency bands is large, the two sets of filters operate independently. When the distance is close, they are merged into one set for coverage. The two sets of signal components are labeled with grouping and enter subsequent processing, thereby completing the cancellation of the corresponding frequency bands at the operating station.

[0061] This embodiment sets up multiple bandpass filters according to the target frequency components. The passbands of each group do not overlap. The signal components of each frequency band are extracted independently and carry group identifiers. The coefficient update is based on the group identifier to locate the corresponding frequency band. If the frequency band spacing is too small, they are automatically merged. The group identifier runs through the entire processing, and the frequency band assignment is clear. Bandwidth constraints and group identifiers together form the basis of frequency band management, thereby avoiding passband compression while maintaining independent processing of each frequency band. The noise cancellation process of each frequency band at the operating station does not interfere with each other. The noise frequency range in the workshop is wide (tens of Hz to thousands of Hz), and the sampling rate requirement is high. If full-band convolution operation is performed on each sampling point, the load on the digital signal processor will be extremely high, resulting in increased system latency and failing to meet the requirements of real-time noise reduction. Moreover, the latency in industrial sites is required to be extremely low (millisecond level). Otherwise, the noise immunity cannot cancel the original noise, and general-purpose processors cannot process multi-channel, high-sampling-rate real-time data simultaneously. This application decomposes long-sequence convolution into short-sequence operations, which greatly reduces the amount of single operation. By filtering the target frequency, invalid calculations are reduced.

[0062] To cover the wide-band noise (e.g., 50Hz-2kHz) of an automotive manufacturing plant, traditional solutions require high-order FIR filters, which places extremely high demands on DSP computing power. Technicians often face a dilemma: either sacrifice the sampling rate and filter order (resulting in insufficient noise reduction depth) or upgrade to expensive high-performance hardware (leading to a surge in costs). This invention, through the synergistic effect of spectral sparsity filtering and multiple sets of narrowband parallel filters, processes only a few target frequency components with concentrated energy. Surprisingly, this "dimensionality reduction" not only reduces the computational load by 1-2 orders of magnitude, enabling the system to run smoothly on low-power embedded platforms, but also significantly improves the signal-to-noise ratio and convergence accuracy of the target frequency band by avoiding interference from irrelevant frequency bands. Combined with the segmented convolution and overlapping addition method mentioned in this application, the system significantly reduces latency while unexpectedly reducing spectral leakage and the picket fence effect, resulting in an output signal purity far exceeding that of full-band filtering schemes. This simplification of complexity, reducing computational power while improving control accuracy, defies conventional technical understanding and has unexpected effects.

[0063] In one embodiment, in step 3, the preset transfer function library is divided into a press transfer function group, a welding robot transfer function group, and a conveyor line transfer function group. The secondary channel transfer functions of each frequency within the press transfer function group are measured at the operating station when the press is running alone. Similarly, the secondary channel transfer functions of each frequency within the welding robot transfer function group are measured at the operating station when the welding robot is running alone. The secondary channel transfer functions of each frequency within the conveyor line transfer function group are measured at the operating station when the conveyor line is running alone. During measurement, an excitation signal is applied to the noise source, and an error microphone receives the response at the operating station. The ratio of the excitation to the response is the secondary channel transfer function for the corresponding frequency. The noise source remains under constant operating conditions during the measurement process. The response signal from the error microphone at the operating station is collected synchronously with the excitation signal. Single-source measurement is used to eliminate interference from the sound fields of other equipment, thereby obtaining the independent sound propagation path characteristics of each noise source. The measurement is performed after the runtime of the corresponding noise source exceeds a preset measurement duration threshold, such as 5 minutes or 10 minutes. The noise source output is minimized, and measurements are performed only after the noise source output has stabilized. The measured secondary channel transfer function, along with the temperature value during measurement, is written into a pre-set transfer function library. Entries in the library are indexed by noise source type and frequency. The transfer function library is stored in the memory of the noise reduction control unit, with its capacity planned according to the number of noise source types and frequency points. Based on the correspondence between the frequency band of the target frequency component and the frequency bands corresponding to each transfer function group, the secondary channel transfer function corresponding to the target frequency component is selected from the corresponding transfer function group. That is, if the target frequency component falls in the press frequency band, it is selected from the press transfer function group; if it falls in the welding robot frequency band, it is selected from the welding robot transfer function group; and if it falls in the conveyor line frequency band, it is selected from the conveyor line transfer function group. The correspondence is pre-established in the library, and the selection result is the candidate value of the current secondary channel transfer function, thus ensuring that the selected acoustic path characteristics are consistent with the current noise source. The selection operation takes the target frequency component as input and outputs the candidate transfer function from the corresponding transfer function group. When the target frequency component simultaneously falls in the frequency bands corresponding to multiple transfer function groups, the summation satisfies... That is, H is the transfer function of the current secondary channel. Let J be the secondary channel transfer function for the corresponding frequency band in the j-th transfer function group, where J is the number of transfer function groups. When multiple devices operate simultaneously, the acoustic path at the operating station is the superposition of the paths from each device. Summation is used to synthesize the equivalent propagation characteristics under simultaneous multi-source action. The summation result is used as the current secondary channel transfer function in this iteration. The number of groups participating in the summation is determined by the number of frequency bands the target frequency component falls into. Each frequency secondary channel transfer function in the preset transfer function library includes the temperature value within the operating station during measurement. The temperature value is collected by a temperature sensor installed at the operating station, positioned in the air near the operating station. The sampling period corresponds to the iteration period. The temperature change is calculated as the difference between the current temperature and the temperature at the time of measurement. When the temperature change within the operating station exceeds a preset temperature threshold, correction is performed. ,Right now H is the modified transfer function of the current secondary channel. The temperature correction coefficient is a preset temperature threshold, such as 2℃ or 3℃. The correction coefficient is calculated based on the temperature deviation according to a preset relationship. Air density changes with temperature, affecting sound velocity and sound propagation loss. The correction coefficient is used to adjust the transfer function amplitude according to the temperature deviation, thereby maintaining the accuracy of path characteristics under different seasons or time periods of workshop temperature changes. Taking a summer afternoon workshop temperature rise as an example, if the temperature at which the transfer function was measured differs from the current temperature by more than the preset temperature threshold, the control unit corrects the current secondary channel transfer function according to the temperature correction coefficient, thus ensuring that the cancellation effect at the operating station does not drift with temperature.

[0064] In this embodiment, the transfer function is measured and stored in groups according to the noise source. When selected, it corresponds to the frequency band where the target frequency component is located. When multiple sources are running simultaneously, the equivalent path is synthesized by summing the groups. When the temperature change exceeds the preset temperature threshold, the transfer function is adjusted according to the correction coefficient. The temperature correction is based on the historical temperature records attached to the library. No additional measuring equipment is required. The measurement and correction processes can be completed during the normal operation of the workshop, thereby maintaining the accuracy of the path description under different equipment combinations and different ambient temperatures.

[0065] The workshop temperature fluctuates drastically with equipment operation, altering the sound velocity and causing the preset filter coefficients to malfunction, resulting in temperature drift. If a single, fixed model is used, the noise reduction effect will be significantly reduced if the workshop layout is slightly adjusted (e.g., by piling up clutter). This application introduces temperature monitoring and correction coefficients, updating the coefficients of the secondary channel transfer function in real time based on error signals, enabling the system to have "self-learning" capabilities and update online.

[0066] The ambient temperature in the workshop fluctuates drastically with equipment operation (temperature differences can reach 10-20℃), causing changes in sound velocity and acoustic path. Existing technologies typically rely on manual recalibration during periodic shutdowns or complex online system identification (which is highly susceptible to noise interference and divergence). Technicians generally believe that maintaining the long-term accuracy of the transfer function in uncontrolled industrial environments is virtually impossible. This invention creatively introduces a temperature correction coefficient, dynamically binding thermodynamic parameters to the acoustic transfer function. The system can automatically compensate for acoustic characteristic drift caused by temperature without shutdown. More surprisingly, by combining precise invocation of a preset transfer function library with micro-correction based on error signals, the system achieves dual stability of "coarse adjustment + fine adjustment." Even under strong airflow and drastic temperature changes, the system maintains extremely high stability, avoiding the "whistling" and divergence phenomena common in traditional ANC systems.

[0067] In one embodiment, in step 4, a piecewise convolution operation is performed on the signal component and the current secondary channel transfer function. The piecewise convolution takes the signal component as input, divides the processing intervals according to the window length, and each interval independently performs convolution with the current secondary channel transfer function. The outputs are arranged in interval order. The length of each segment in the piecewise convolution operation is the same as the window length of the Hanning window. Segment processing starts from the beginning position of the signal component, and the result of each segment is output immediately after processing, thereby reducing the amount of intermediate data remaining. The interval between the starting positions of two adjacent segments is half the window length, meaning there is an overlap of half the window length between adjacent segments. This overlap design is used to avoid discontinuities in the convolution results at the segment boundaries, thus ensuring smooth transition of the output secondary noise signal between segments. The convolution operation in the piecewise convolution operation satisfies... ,in Let be the secondary noise signal at time t. This is the m-th coefficient of the current secondary channel transfer function. Let be the value of the signal component at time tm, and L be the number of coefficients in the current secondary channel transfer function. The convolution operation weights and superimposes the signal components point-by-point according to the impulse response of the acoustic path, simulating the amplitude and phase state of the secondary sound wave arriving at the error microphone position. The outputs of the segmented convolution operation are combined in an overlapping addition manner. In this overlapping addition method, the outputs of two adjacent segments are averaged within the overlapping interval and then added. The average result is then concatenated with the outputs of the remaining intervals in time sequence. The average is used to eliminate the small differences between two independent calculations within the overlapping interval, thus obtaining a continuous secondary noise signal. The length of the overlapping interval is related to the interval between the starting positions of adjacent segments; when the interval is half the window length, the length of the overlapping interval is also half the window length. The sampling frequency of the digital-to-analog conversion satisfies... The sampling frequency of the digital-to-analog converter (DAC) must be no less than twice the frequency value of the maximum target frequency component. For example, the sampling frequency can be 2.5 or 3 times the maximum target frequency component. This sampling frequency constraint ensures that the highest target frequency component does not aliasing during DAC, thus preserving the complete spectral information of the secondary noise signal. When the sampling frequency matches the preset sampling frequency, DAC and data acquisition share the same clock source, maintaining a unified time base. When the peak amplitude of the secondary noise signal exceeds the rated input amplitude of the speaker, the peak amplitude is compressed to the rated input amplitude of the speaker, resulting in a compressed secondary noise signal. DAC is then performed on the compressed secondary noise signal to obtain an analog drive signal. Compression is performed before DAC, ensuring the compressed signal amplitude no longer exceeds the rated input amplitude. Peak amplitude detection is completed before compression, with the detection window covering the entire secondary noise signal. Compression processing prevents the drive signal from exceeding the speaker's allowable input range, thus preventing speaker overload distortion. The cross-correlation function between the secondary noise signal and the error noise signal is calculated from the sampled values ​​of the secondary noise signal and the error noise signal within a preset time period. The preset time period is the same as the window length of the Hanning window. The cross-correlation calculation slides point by point within the preset time period to measure the similarity between the secondary noise signal and the residual noise. Its peak value reflects the phase alignment between the secondary sound wave and the noise source sound wave at the position of the error microphone. The peak value is compared with a preset correlation threshold to determine whether to replace the transfer function. If the peak value of the cross-correlation function is not greater than the preset correlation threshold, the current secondary channel transfer function is replaced with the corresponding secondary channel transfer function in the preset transfer function library, and step 6 is continued. If the peak value is too low, it indicates that the current transfer function deviates from the actual acoustic path. It is replaced with a pre-determined value in the library to re-establish an accurate path description, thereby preventing the control process from continuously accumulating errors on the wrong path. Taking a small workstation with limited speaker output power as an example, the limiting processing ensures that the drive signal is always within the rated range of the speaker, thereby maintaining the continuous output of the secondary sound wave while ensuring safe operation.

[0068] This embodiment uses a combination of segmented convolution and overlapping addition for output. The length of each segment is constrained by the window length. The operation process proceeds segment by segment, and the data usage decreases accordingly. The sampling frequency is constrained by the highest target frequency to avoid aliasing. When the peak value exceeds the limit, it is compressed to the rated range. When the cross-correlation peak value is too low, the current transfer function is replaced by the measured value in the library. The limiting and replacement mechanism form a double guarantee at the hardware protection level. Segmented processing makes the processing of long signals more stable, thus forming a closed loop between the operation scale, speaker safety and path deviation correction.

[0069] In one embodiment, in step 6, the root mean square value of the error noise signal within a preset time period satisfies Where RMS is the root mean square value of the error noise signal over a preset time period. Let K be the k-th sampled value of the error noise signal, where K is the number of samples of the error noise signal within a preset time period. The preset time period is the same as the window length of the Hanning window. The root mean square (RMS) value is calculated using a sliding window method, with the window length being the same as the Hanning window length. The window moves forward gradually with each iteration. The sampled values ​​within the window are first squared point by point and then summed. The summation result is divided by the number of samples and then the square root is taken. The RMS value is used to characterize the average energy level of the residual noise within the preset time period, thereby avoiding fluctuation interference caused by using single-point amplitude as the evaluation criterion. The updated difference satisfies The update difference is calculated as the difference between the root mean square (RMS) value and the preset target noise value. The difference is calculated using the RMS value and the preset target noise value as inputs. A negative difference indicates that the current residual noise is below the expected level. The difference is calculated at the beginning of each iteration, with the iteration period corresponding to a preset duration. The preset target noise value corresponds to the expected residual noise level at the operating station, for example, 45 dB or 50 dB. When the absolute value of the update difference is not greater than a preset update threshold, the coefficients of the current secondary channel transfer function remain unchanged. The preset update threshold is, for example, 1 dB or 2 dB. A small difference indicates that the current cancellation effect is close to the expected level; maintaining the coefficients avoids frequent adjustments caused by small fluctuations. When the absolute value of the update difference is greater than the preset update threshold, the coefficients corresponding to the target frequency component in the current secondary channel transfer function are updated according to a preset step size. The preset step size satisfies... The preset step size is the product of the preset base step size and the absolute value of the update difference. The preset base step size is, for example, 0.01 or 0.02. The larger the difference, the larger the step size, thus accelerating convergence when the deviation is large and reducing the adjustment amplitude when approaching the desired level. The preset base step size is determined during the debugging phase; a value that is too small results in slow convergence, while a value that is too large is prone to oscillation. The update is completed within the current iteration, and the updated coefficients are immediately used for the convolution operation in the next iteration. Coefficients in the updated secondary channel transfer function that do not correspond to the target frequency component remain at their pre-update values. That is, only the coefficients of the current effective frequency band are adjusted, while other frequency bands remain unaffected, thus avoiding the co-current disturbance of frequency bands not involved in cancellation. Error noise signals are continuously acquired during the coefficient holding period, and the update automatically resumes according to the step size after the difference falls back. When the absolute value of the update difference exceeds the preset update threshold for a certain number of consecutive times, reaching the preset upper limit, the coefficient corresponding to the target frequency component in the current secondary channel transfer function is multiplied by a preset attenuation coefficient. The preset attenuation coefficient is not less than 0.5 and not greater than 0.9. The preset upper limit is, for example, 20 or 30. The attenuation coefficient is applied to the coefficients corresponding to the target frequency component. After multiplying the coefficients by the attenuation coefficient, the amplitude decreases. Consecutive deviations indicate that the current path may have shifted or the gain is too large. The attenuation coefficient is used to reduce the overall coefficient amplitude, thereby suppressing the divergence trend. After attenuation, the coefficient amplitude decreases, and the next iteration readjusts the coefficients based on the difference. When the root mean square value is not less than twice the preset target noise value, the coefficient update for this iteration is skipped, and the updated secondary channel transfer function is kept as the current secondary channel transfer function before the update. When the residual noise reaches more than twice the expected level, the adjustment is temporarily stopped, and the error noise signal continues to be collected during the skipped update period. When the root mean square value falls back to within twice the preset target noise value, the coefficient update is resumed, thereby avoiding blindly increasing the coefficients under abnormal operating conditions, which could lead to control instability. Taking the scenario of continuous high-speed operation of a stamping press as an example, when the residual noise exceeds twice the preset target noise value, the control unit keeps the coefficient unchanged. After the noise drops back to the normal range, the coefficient is updated, thereby maintaining the stability of the control process during equipment malfunctions.

[0070] In this embodiment, the root mean square value within a preset time period is used to measure the residual noise energy. The step size adapts to the absolute value of the update difference. When the difference is within the preset update threshold, the coefficient remains unchanged. When the difference exceeds the limit continuously, the coefficient amplitude is reduced by the attenuation coefficient. When the residual noise exceeds twice the expected level, the adjustment is temporarily stopped. The attenuation and pause mechanism is used to deal with abnormal operating conditions, thereby achieving a balance between convergence speed and stability. The control process remains stable under abnormal operating conditions.

[0071] The active noise control system in the complex acoustic environment of an automobile manufacturing workshop is a feedback control system, which is prone to self-excited oscillation (howling) or algorithm divergence (the noise increases with noise reduction). When there are human voices or other sudden sounds in addition to equipment noise in the workshop, the system may mistakenly amplify these sounds as residual noise. The fixed-step LMS algorithm converges slowly when the error is large and is unstable when the error is small. This application dynamically adjusts the update step size according to the error magnitude, achieving fast convergence with large errors and maintaining stability with small errors; it uses a cross-correlation function to determine the validity of the signal, preventing the system from responding incorrectly to irrelevant signals; and it automatically attenuates the coefficient when the error remains too large to prevent system collapse.

[0072] In one embodiment, in step 5, the loudspeaker consists of multiple loudspeaker units, for example, 4 or 6 loudspeaker units. Each unit is connected to the noise reduction control unit via an independent power amplification channel. All loudspeaker units have the same model and the same rated input amplitude. The multiple loudspeaker units are arranged around the operating station. The distance between the multiple loudspeaker units and the error microphone is no greater than a third preset distance threshold, for example, 0.5 m or 0.8 m. The angle between any two adjacent loudspeaker units and the center of the operating station is the same. The center of the operating station is the geometric center of the operating station. The arrangement around the loudspeaker units and the uniform angle are used to make the spatial positions of each loudspeaker unit symmetrically distributed to the operating station, thereby providing a consistent geometric basis for sound wave superposition. The analog drive signals are amplitude-distributed according to the distance between each loudspeaker and the error microphone. Loudspeakers at greater distances receive larger amplitudes, and those at smaller distances receive smaller amplitudes. The greater the distance, the greater the sound wave attenuation. Amplitude distribution is based on distance, and the amplitude allocated to each speaker is directly proportional to the distance. The distribution is applied before the delay. Loudspeakers at greater distances receive larger drive amplitudes. Amplitude distribution is used to compensate for propagation loss at different distances, thereby making the sound pressure generated by each loudspeaker at the error microphone position more consistent. The distributed analog drive signals are then delayed according to the delay amount corresponding to the distance between each loudspeaker and the error microphone before being input to the corresponding loudspeaker. Each delay amount satisfies… ,Right now Let be the delay of the i-th speaker unit. Let be the distance between the i-th loudspeaker unit and the error microphone, and c be the preset sound velocity, which is 340 m / s. The delay is used to align the arrival time of the sound waves from each loudspeaker unit at the error microphone position. Units farther away emit sound earlier, so that the sound waves from all paths arrive at the error microphone position at the same time and are superimposed in phase. The delay is arranged in ascending order of distance, with smaller delays for units with smaller distances. The peak amplitude of each analog drive signal after delay is not greater than the rated input amplitude of each loudspeaker unit, thus ensuring that each loudspeaker unit operates within its rated operating range. The orientation of each loudspeaker unit points towards the center of the operating position. Orientation adjustment is completed during the installation phase. After installation, the main radiation direction of each unit is aligned with the center of the operating position. Each unit is installed on the columns or brackets around the operating position, with the installation height level with the operator's ear. The orientation setting is used to align the directional main lobe of the loudspeaker with the operating position, thereby reducing the radiation of sound energy in irrelevant directions. Taking a scenario where four speaker units are evenly arranged around the workstation as an example, each unit is assigned an amplitude according to the distance and a corresponding delay is applied. The secondary sound waves are superimposed in phase at the operator's ear position, thereby forming a relatively uniform canceling sound field at the error microphone position.

[0073] In this embodiment, multiple loudspeaker units are arranged around the operating station. The distance between each unit and the error microphone is limited. The amplitude is compensated for the propagation loss according to the distance, and the delay is aligned with the arrival time of the sound speed according to the distance. The orientation is towards the center of the operating station, and the sound field distribution at the ear position tends to be uniform. The amplitude distribution and delay alignment are both based on distance as input. The implementation method is simple. The geometric structure of the surrounding arrangement also provides symmetry for the sound field superposition. The cancellation effect can still remain stable when the operator moves his head slightly.

[0074] The workstation is a three-dimensional space, while the worker's ear position is dynamic. A single speaker can hardly create a stable, silent zone over a large area. Due to sound wave interference, noise reduction might be successful at point A but amplified at point B. The position of the error microphone deviates from the worker's actual ear position, resulting in a perceived weak noise reduction effect. This application uses multiple speakers arranged around the workstation to create surround noise reduction. The output amplitude and delay (phase calibration) are adjusted based on the distance from the speakers to the microphone to ensure that all sound waves superimpose and cancel each other out at the worker's ear position.

[0075] This embodiment creates a silent zone with "spatial resilience," overcoming the limitations of point noise reduction in strong reverberation environments. Workshops often have hard reflective surfaces with extremely long reverberation times. Traditional single-channel active noise control systems can only achieve noise reduction at the "point" where the error microphone is located. Even slight head movements by operators drastically worsen the noise reduction effect (a so-called "point noise reduction trap"). Technicians typically believe that constructing a "silent zone" of a certain volume in an open workshop requires extremely complex beamforming algorithms and a large number of loudspeakers, making the cost prohibitive. This invention, through a multi-speaker ring array combined with distance-amplitude-delay joint control, constructs a stable "silent body" rather than a "silent point" at the workstation. Unexpectedly, the introduced differential signal monitoring mechanism allows the system to sense subtle changes in the noise source and fine-tune the output phase of each loudspeaker in real time. This dynamic spatial sound field reconstruction capability gives the silent zone resilience—even with worker movement or temporary obstruction by workpieces, the noise reduction effect does not drop drastically. This robust spatial noise reduction effect in strong reverberation environments exceeds the expectations of those skilled in the art and has unexpected benefits.

[0076] In one embodiment, in step 4, each signal component is multiplied by a Hanning window before the convolution operation. The windowing operation is performed before each signal component is input into the convolution. The window function is the same as the Hanning window used in the Fourier transform stage, that is, each data segment entering the convolution is first weighted by the window function to suppress abrupt changes at the segment edges. The length of the overlap interval between two adjacent signal components is half the window length, thereby ensuring that adjacent segments after windowing can be smoothly connected in the overlap region. The quantization bit depth of the digital-to-analog conversion is not less than 16 and not more than 24. For example, the quantization bit depth is 16 bits or 24 bits. The quantization bit depth is determined by the dynamic range requirements of the analog driving signal. When the bit depth is higher, the quantization noise has less impact on the signal. The choice of quantization bit depth matches the sampling frequency. When the bit depth is higher, the amount of data in the digital-to-analog conversion increases accordingly. The quantization bit depth determines the amplitude resolution of the analog driving signal, thereby maintaining a small quantization error within the signal's dynamic range. When the peak amplitude of the secondary noise signal exceeds the preset peak threshold, the peak amplitude of the secondary noise signal is multiplied by a preset compression coefficient. The preset compression coefficient is not less than 0.5 and not more than 0.9. The preset peak threshold is a preset ratio of the rated input amplitude of the speaker. The preset ratio is not less than 0.8. That is, when the peak value is close to the rated input amplitude of the speaker, the signal is compressed. The compression coefficient is applied to the peak amplitude. After the peak amplitude is multiplied by the compression coefficient, it does not exceed the rated input amplitude. The compression coefficient pulls the peak value back within the rated range, thereby preventing signal overload from damaging the speaker. When the amplitude of the secondary noise signal does not exceed the preset lower amplitude limit within a preset duration, the analog drive signal is set to a zero amplitude signal. The preset lower amplitude limit is one-hundredth of the speaker's rated input amplitude, and the preset duration is the same as the window length of the Hanning window. The silent state is triggered by continuous low amplitudes. During the silent period, all analog drive signals are set to zero. The duration of the zero amplitude signal does not exceed the preset upper limit of the silent output duration, such as 2 seconds or 5 seconds. Continuous low amplitudes indicate that there are currently no effective noise components that need to be canceled. The zero amplitude signal is output to avoid the speaker continuously generating unnecessary current in a silent state. The upper limit of the silent duration is used to restore the output in a timely manner after the noise is restored. Zero-filling interpolation is performed on the analog drive signal, and the interpolated sampling frequency meets the following requirements. ,Right now The sampling frequency after zero-padding interpolation. The sampling frequency for digital-to-analog conversion is set to 2, with an interpolation factor of 2. Zero-fill interpolation inserts zero values ​​between adjacent sampling points and performs low-pass smoothing to increase the sampling density of the analog drive signal. Interpolation is performed after digital-to-analog conversion. The interpolated sampling sequence is truncated or padded according to the number of sampling points corresponding to the window length, thus providing a finer time resolution for subsequent delay alignment and speaker driving. The number of sampling points of the analog drive signal after zero-fill interpolation is the same as the number of sampling points corresponding to the window length of the Hanning window. The sampling point constraint is used to ensure that the length of the interpolated signal matches the window structure, thereby maintaining the consistency of the time base for segmented processing. Taking the speaker standby state as an example, when there is no noise source at the operating station, the amplitude of the secondary noise signal is continuously below the lower limit, and the control unit outputs a zero-amplitude signal. After the noise source restarts, the analog drive signal output is restored. After the silence is restored, the analog drive signal smoothly rises from zero amplitude. The rise process is segmented and gradually transitioned according to the window length, thereby reducing the static power consumption of the speaker while maintaining the system response.

[0077] In this embodiment, a Hanning window is applied to each signal component before convolution, the number of quantization bits is selected within a certain range, the peak value is limited to the rated range according to the compression factor, zero amplitude signal is output during continuous low amplitude and silenced within the upper limit of duration, and zero-fill interpolation is used to increase the sampling density, thereby achieving a balance between signal accuracy, speaker safety and static power consumption. The interpolation processing is completed on the basis of existing sampling without changing the front-end acquisition structure. The values ​​of each link are all based on speaker safety and signal accuracy, and the control of delay alignment and speaker driving is more refined.

[0078] To counteract high-intensity industrial noise, active noise control systems in the complex acoustic environment of automotive manufacturing plants require high-power output signals. This can easily lead to overheating and burnout of the speaker voice coil, or generate clipping distortion (creating new high-frequency harmonic pollution). Prolonged high dynamic range output can shorten speaker lifespan, and the distortion noise generated by signal overload may be more jarring than the original noise. This application addresses this by compressing the signal before digital-to-analog conversion to ensure it does not exceed the speaker's rated input amplitude; when the signal is detected to be too low, a zero signal is output to allow the speaker to rest.

[0079] To counteract high-intensity industrial noise, active noise control systems often drive loudspeakers at full load for extended periods, leading to overheating and burnout of the voice coil or fatigue damage to the surround. Technicians typically believe that noise reduction performance and hardware lifespan are contradictory; improving performance inevitably sacrifices lifespan. This invention, through amplitude compression, zero-fill interpolation, and silent detection, not only protects the hardware but also unexpectedly discovers a positive feedback loop of "noise reduction-protection." When the system detects that the signal amplitude is too low (e.g., when equipment stops), it automatically outputs a zero signal (silent mode), saving energy and cooling the loudspeaker. This intelligent intermittent working mechanism, combined with a compression algorithm that prevents clipping distortion, extends the loudspeaker's lifespan several times over while avoiding new high-frequency harmonic pollution caused by hardware distortion. This "side effect" of transforming algorithm optimization into improved hardware reliability is something that equipment maintenance personnel did not anticipate.

[0080] In one embodiment, such as Figure 2 As shown, this application also provides a multi-source adaptive active noise control system for automobile manufacturing workshops, including a reference microphone group, a beat sensor, an error microphone, a speaker array, and a signal processing device. The reference microphone group is arranged near noise sources such as stamping machines, welding robots, or conveyor lines within the automobile manufacturing workshop. Each reference microphone group contains multiple individual reference microphones evenly arranged circumferentially along the noise source for acquiring reference noise signals. The reference microphone group is connected to the signal processing device via shielded cables. The acquisition channels are synchronously triggered at a preset sampling frequency, thereby ensuring that the reference signals are aligned in time. The beat sensor is connected to the stamping machine to acquire the stamping beat signal. The stamping beat signal is input to the signal processing device in the form of level transitions to indicate the timing of the stamping stroke, providing a trigger reference for the extraction of impact noise signal segments. An error microphone is positioned at the workstation to collect the error noise signal after noise reduction processing. The microphone is mounted on a bracket near the operator's ear, with the distance between it and the ear not exceeding a second preset distance threshold, ensuring that the collected residual noise closely approximates the operator's actual perception. A speaker array is arranged around the workstation, consisting of multiple identical speaker units with the same rated input amplitude. Each speaker unit is connected to a signal processing device via an independent power amplification channel to convert the analog drive signal into a secondary sound wave output to the workstation. The angle between any two adjacent speaker units and the center of the workstation is the same, and all speaker units point towards the center of the workstation.

[0081] The signal processing unit is connected to the reference microphone group, the beat sensor, the error microphone, and the loudspeaker array to execute the above method. Specifically, the signal processing unit includes a signal extraction unit, a spectrum extraction unit, a transfer function matching unit, a filtering and convolution operation unit, a digital-to-analog conversion and compression unit, an amplitude allocation and delay compensation unit, and a coefficient update unit. The signal extraction unit extracts the impact noise signal segment from the reference noise signal based on the rising edge of the impact beat signal output by the beat sensor. The spectrum extraction unit applies a Hanning window to the reference noise signal and performs a Fourier transform, marking the target frequency component with an amplitude greater than a preset threshold. The transfer function matching unit selects the corresponding current secondary channel transfer function from a preset transfer function library based on the frequency band where the target frequency component is located, and performs temperature correction based on the temperature value at the operating station. The filtering and convolution operation unit uses a bandpass filter to extract the signal components and then processes the signal components... The secondary noise signal is obtained by convolving it with the current secondary channel transfer function. The analog-to-digital conversion and compression unit compresses the peak amplitude of the secondary noise signal before the digital-to-analog conversion to ensure that it is not greater than the rated input amplitude of the loudspeaker, thus obtaining the analog drive signal. The amplitude allocation and delay compensation unit performs amplitude allocation and delay compensation on the analog drive signal according to the distance between each loudspeaker unit and the error microphone, and then inputs it to each loudspeaker unit. The coefficient update unit updates the coefficients corresponding to the target frequency component in the current secondary channel transfer function according to the difference between the error noise signal and the preset target noise value, and uses the updated secondary channel transfer function as the current secondary channel transfer function for the next iteration, thereby forming a closed-loop control based on the actual residual noise at the operating station within the signal processing device.

[0082] In this embodiment, the reference microphone group, the beat sensor, the error microphone, the speaker array, and the signal processing device together constitute a complete active noise control system. Each component corresponds one-to-one with the steps of the aforementioned method embodiment. The system can automatically match the corresponding transfer function group and update the control coefficients during the start-up, shutdown, and beat switching of equipment such as stamping machines, welding robots, and conveyor lines without manual intervention. This enables real-time noise reduction with low latency, high stability, and high precision in the high-dynamic, multi-interference, and high-impact environment of automobile manufacturing workshops.

[0083] The unexpected effect of this invention lies in the qualitative leap brought about by synergistic coupling. It is not a linear superposition of single technologies, but rather a deep coupling of signal preprocessing (FFT filtering), physical field linkage (temperature correction), spatial sound field reconstruction (array control), and adaptive algorithms (variable step size updates). This solves the long-standing problems of poor stability, slow convergence, easy divergence, and short lifespan in the field of industrial active noise control. These effects demonstrate that this invention cannot be obtained by those skilled in the art through simple logical reasoning or conventional experiments, and possesses outstanding substantive features and significant progress.

[0084] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. A multi-source adaptive active noise control method for automobile manufacturing workshops, characterized in that, Includes the following steps: Step 1: Divide the automobile manufacturing workshop into multiple noise reduction zones. Place a set of reference microphones near the stamping machine, welding robot or conveyor line in each noise reduction zone, and place error microphones at the operating stations. Obtain the stamping cycle signal of the stamping machine. Based on the rising edge of the stamping cycle signal, extract the impact noise signal segment from the reference noise signal collected by the reference microphone. The time difference between the start time of the impact noise signal segment and the rising edge of the stamping cycle signal is less than a preset time threshold. Step 2: Apply a Hanning window to the reference noise signal and perform a Fourier transform to obtain a frequency domain amplitude sequence. Mark the frequency components in the frequency domain amplitude sequence whose amplitudes are greater than a preset threshold as target frequency components; the preset threshold satisfies... , The threshold value is a preset value, and k is a preset multiple, where k is not less than 2 and not greater than 4. This is the average value of all amplitudes in the frequency domain amplitude sequence; Step 3: Based on the frequency band of the target frequency component, select the corresponding current secondary channel transfer function from the preset transfer function library; the preset transfer function library is divided into a stamping machine transfer function group, a welding robot transfer function group, and a conveyor line transfer function group. The transfer functions of each group are measured at the operating station when the corresponding noise source is running alone; monitor the temperature value in the operating station, and when the change in temperature value is greater than the preset temperature threshold, perform temperature correction on the current secondary channel transfer function based on the physical model of sound speed changing with temperature; Step 4: Use a bandpass filter to filter the reference noise signal to obtain the signal components, and convolve the signal components with the current secondary channel transfer function to obtain the secondary noise signal; perform digital-to-analog conversion on the secondary noise signal, and compress the peak amplitude of the secondary noise signal before conversion to make it no greater than the rated input amplitude of the speakers arranged around the operating station to obtain the analog drive signal; Step 5: Distribute the analog drive signals to the speaker array arranged around the operating station; Based on the distance between each loudspeaker unit and the error microphone located at the operating position, the signal input to each loudspeaker unit is amplitude-distributed and delay-compensated so that the sound waves of each channel are superimposed and canceled at the position of the error microphone. Step 6: Based on the difference between the error noise signal collected by the error microphone and the preset target noise value, update the coefficients corresponding to the target frequency component in the current secondary channel transfer function to obtain the updated secondary channel transfer function, and use it as the current secondary channel transfer function for the next iteration.

2. The method according to claim 1, characterized in that, In step 1, the reference microphone group includes multiple reference microphone units uniformly arranged around the noise source. Differential signal sequences are constructed by subtracting the signals collected by the multiple reference microphone units pairwise. The switching time of the noise source is identified based on the amplitude change of the differential signal sequence, so as to dynamically adjust the interception window of the impact noise signal segment.

3. The method according to claim 1, characterized in that, In step 2, the reference noise signal is divided into multiple signal segments. After performing a Fourier transform on each signal segment, the amplitudes of each signal segment at the same frequency are averaged to obtain the average frequency domain amplitude sequence. The number of target frequency components is counted. When the number of target frequency components is greater than the preset upper limit, the preset upper limit of frequency components are selected as target frequency components in descending order of amplitude. When the target frequency component of the current iteration is exactly the same as the target frequency component of the previous iteration and the number of consecutive iterations reaches the upper limit of the preset number of iterations, the preset multiple k is reduced by the preset adjustment amount.

4. The method according to claim 1, characterized in that, In step 4, multiple bandpass filters are used to filter the reference noise signal. Each bandpass filter corresponds to a target frequency component, and the bandwidth of each bandpass filter is no greater than 1 / 2 of the frequency interval between two adjacent target frequency components. When the frequency interval between two adjacent target frequency components is less than a preset interval threshold, the two adjacent target frequency components are merged into one target frequency component, and the frequency band range of the bandpass filter is re-determined based on the merged target frequency component.

5. The method according to claim 1, characterized in that, In step 3, when the target frequency component is located in the frequency band corresponding to multiple transfer function groups simultaneously, the current secondary channel transfer function H is obtained by summing the following formulas: ； Let J be the secondary channel transfer function of the corresponding frequency band in the j-th transfer function group, where J is the number of transfer function groups.

6. The method according to claim 1, characterized in that, In step 4, a piecewise convolution operation is performed on the signal components and the current secondary channel transfer function. The length of each segment of the piecewise convolution operation is the same as the window length of the Hanning window, and the interval between the starting positions of two adjacent segments is 1 / 2 of the window length. The outputs of the piecewise convolution operation are combined in an overlapping addition manner. When overlapping addition, the outputs of two adjacent segments are averaged in the overlapping interval and then added together to obtain the secondary noise signal.

7. The method according to claim 1, characterized in that, In step 6, the update difference is calculated. , where T is the preset target noise value, and RMS is the root mean square value of the error noise signal within the preset time period; when |e| is greater than the preset update threshold, the coefficient is updated according to the preset step size μ, the preset step size μ=μ0×|e|, where μ0 is the preset base step size; when the number of times |e| is greater than the preset update threshold reaches the preset upper limit, the coefficient corresponding to the target frequency component in the current secondary channel transfer function is multiplied by the preset attenuation coefficient, the preset attenuation coefficient α∈[0.5,0.9].

8. The method according to claim 1, characterized in that, In step 5, the loudspeaker array consists of multiple loudspeaker units, which are evenly arranged around the center of the operating station, with the same angle between any two adjacent loudspeaker units and the center of the operating station; the delay of the i-th loudspeaker unit... satisfy: ； in, Let be the distance between the i-th loudspeaker unit and the error microphone, and c be the preset sound velocity.

9. The method according to claim 1, characterized in that, In step 4, when the amplitude of the secondary noise signal does not exceed the preset lower limit of amplitude within a preset time period, the analog drive signal is set to a zero amplitude signal, and the preset lower limit of amplitude is 1% of the rated input amplitude of the speaker; Zero-fill interpolation is performed on the analog drive signal, with the interpolation factor set to 2, to improve the smoothness of the output signal.

10. A multi-source adaptive active noise control system for automobile manufacturing workshops, characterized in that, include: A reference microphone array is placed near the noise source in the workshop to collect reference noise signals; A beat sensor, connected to the stamping press, is used to acquire the stamping beat signal; An error microphone is placed at the operating station to collect error noise signals. A loudspeaker array, arranged around the workstation, is used to output noise immunity; A signal processing device, respectively connected to a reference microphone array, a beat sensor, an error microphone, and a loudspeaker array, is configured to perform the method as claimed in any one of claims 1-9.

Images