Fan pipeline noise reduction system of double-ring active and passive coupling structure

Abstract

The invention discloses a fan pipeline noise reduction system of a double-loop active and passive coupling structure, and belongs to the field of active noise control. The fan pipeline noise reduction system comprises the following steps that a feedforward-feedback double-loop control structure is constructed; a bilateral hybrid control algorithm of adaptive filtering fusion is adopted; introducing a momentum item factor; the beneficial effects of the invention lie in that through the synergistic effect of the inner and outer double-ring structures, the real-time suppression capability of broadband and narrowband composite noise is effectively improved; the adaptive filtering fusion mechanism enhances the adaptability and robustness of the system to the non-stationary noise environment of the fan pipeline; the introduction of the momentum factor further accelerates the convergence speed of the system, and significantly reduces the steady-state residual error level. Compared with a traditional single-loop control structure, the system shows better stability and noise reduction performance in a nonlinear complex sound field, and is suitable for efficient and accurate fan pipeline noise treatment in an industrial scene.

Description

TECHNICAL FIELD

[0001] The fan pipeline noise reduction system with a double-ring active-passive coupling structure belongs to the field of active noise control. BACKGROUND

[0002] Active noise control (ANC) is an advanced low-frequency noise suppression technology based on the principle of destructive interference. It has been widely studied since the 1980s and has been successfully applied in headphones, air conditioners, rotating machinery, vehicles, and aviation. The structure of the ANC system mainly includes feedforward (FFANC), feedback (FBANC), and hybrid (HANC), among which the feedforward type can be further divided into wideband (FFBANC) and narrowband (FFNANC) control. The traditional FFBANC has limitations in convergence speed and stability when dealing with noise containing wideband and narrowband components, making it difficult to meet the demand for fast response.

[0003] To overcome this bottleneck, researchers have proposed various hybrid ANC schemes, such as introducing sinusoidal modeling, linear prediction filters (LPF), or control structures based on speed sensors. However, these methods often come with high computational complexity or difficulty in parameter adjustment. In recent years, fan-pipeline systems have become the focus of ANC application research due to their multiple noise characteristics such as turbulence, resonance, and blade passage frequency (BPF). Existing work has attempted to use speed measurement, singular spectrum analysis, or multi-channel feedback structures to achieve active noise reduction for fans, but these methods are generally limited to the BPF frequency band and lack effective control over wideband and resonant noise.

[0004] In China, ANC technology research started relatively late, and applications are mostly concentrated in headphones, trains, helicopters, and other scenarios. Few studies have been conducted on fan-pipeline systems, and most are still in the simulation stage, lacking engineering verification. Existing fan noise reduction methods mostly use passive structures, which have problems such as large size, high cost, poor low-frequency control effect, and are not conducive to system efficiency improvement.

[0005] This invention focuses on the wideband composite aerodynamic noise control in industrial high-ventilation pipeline fans. To address the problems of low-frequency suppression in traditional passive methods and poor stability in active control systems in strong turbulent environments, research will be conducted from the aspects of multi-field coupling modeling, hybrid algorithm optimization, and system integration. The goal is to build an integrated noise reduction system with the functions of "intelligent sensing - dynamic regulation - interference dissipation", providing efficient and intelligent noise control solutions for new infrastructure fields such as data centers and industrial plants. SUMMARY

[0006] Aiming at the existing pipeline fan noise reduction research, aiming at the future research direction, the application provides a fan pipeline noise reduction system with double-ring active-passive coupling structure to realize the following purposes: effectively improving the real-time suppression ability of wideband and narrowband composite noise; enhancing the adaptability and robustness of the system to the non-stationary noise environment of the fan pipeline; further accelerating the convergence speed of the system and significantly reducing the steady-state residual level.

[0007] The purpose of the application is achieved as follows:

[0008] Step a, using double-layer pipeline structure, the inner pipeline wraps the fan to realize the preliminary reduction of mechanical noise. At the same time, sensors are arranged on both sides of the fan to effectively control the aerodynamic noise. The inner ring adopts a parameter self-tuning notch filter array to eliminate mechanical harmonics, and the outer ring improves the FxLMS algorithm to suppress wideband aerodynamic noise.

[0009] Step b, for aerodynamic noise, a new double-sided feedforward hybrid active noise control structure is adopted, an additional microphone is added on the air inlet side of the outer pipeline fan to capture another reference signal containing high-power blade passing frequency components and low-power turbulent noise, which greatly reduces the difficulty of extracting narrowband components. In addition, the system is equipped with an SNC for extracting the wideband component in the reference signal x r (n) obtained on the exhaust side, the input of the SNC is a sine wave with a small amount of additive noise. The input and error signals are calculated as:

[0010] e f,k (n)=h k,0 (n)x f,k (n)+h k,1 (n)x f,k (n-1),k=1,2,...q

[0011]

[0012] Wherein, is the MPA coefficient in the SNC. Each MPA uses the LMS algorithm to update:

[0013] h k,0 (n+1)=h k,0 (n)+μ SNC,k x ω (n)x f,k (n)

[0014] h k,1 (n+1)=h k,1 (n)+μ SNC,k x ω (n)x f,k (n-1)

[0015] where μ SNC,k is the step size for updating, f,k f,0 (n) will converge to a single sinusoid, which is the kth sinusoid in the narrowband component x f,0 (n).

[0016] The error signal x ω (n) from the SNC contains the wideband components and is used as input to the sub-controllers W B (z). B (z) is updated using the FXLMS algorithm:

[0017]

[0018] where μ B is the step size for updating, and denote the length of the filter and the filter coefficients, respectively, are the estimates of the impulse response coefficients of the secondary path S Z with length M.

[0019] As in the SNC, controlling each frequency component harmonic individually improves the prediction accuracy of its amplitude and phase, resulting in better stability and control performance. The coefficient output of each sub-controller is calculated as follows:

[0020] y f,k (n) = ω N,k,0 (n) x f,k (n) + ω N,k,1 (n) x f,k (n - 1), k = 1, 2,..., q

[0021] Unlike the coefficients in the SNC, each coefficient in the sub-controllers is updated using the FXLMS algorithm:

[0022] ω N,k,0 (n + 1) = ω N,k,0 (n) + μ N,k e(n) x f,k (n)

[0023]

[0024] where μ N,k denotes another step size and e(n) denotes the residual error. Finally, the secondary source is used to cancel the loudspeaker emissions to reduce the noise at the residual error microphone location. The auxiliary sources y(n), y p (n), e(n), are calculated as follows:

[0025] y(n) = yω (n) + y f (n)

[0026]

[0027] e(n) = p(n) - y p (n)

[0028] Step c, for motor noise, a momentum term factor is introduced in the FLANN system. At this time, the output signal of the FIR filter is:

[0029]

[0030] FLANN performs nonlinear orthogonal expansion and weighted accumulation on the reference input signal through multiple linear combiners. Let the reference input signal be x(n), represents a first-order delay operation, represents the weight coefficient of the linear combiner, and the nonlinear orthogonal expansion of the p-th order is performed on the input vector of length N, to obtain:

[0031]

[0032] where p represents the order of the nonlinear expansion term, p ∈ ([1, P1]), and P1 is the maximum value of the order; [·] T represents the transposition operation of the vector. In this case, the output signal of the FLANN system composed of P1 linear combiners is:

[0033]

[0034] The output signal of the adaptive active noise control system based on FLANN is composed of the FIR output signal y fir (n) and the FLANN processing signal y f (n), and the specific expression is:

[0035] y(n) = y f (n) + y fir (n)

[0036] The error signal is expressed as:

[0037] e fla (n) = p(n) - y fla (n)

[0038] The LMS algorithm is used to dynamically adjust the coefficients of the FLANN filter:

[0039] α p,i (n+1) = α p,i (n) - μ fla e fla (n) sin[pπx(n-i)]

[0040] beta p,i (n+1) = beta p,i (n) - mu fla e fla (n)cos[p pi x(n-i)]

[0041] where mu fla is the update step of the FLANN filter coefficient weight.

[0042] The cost function of the M-FLANN algorithm based on the momentum term technique is constructed as follows:

[0043]

[0044] where lambda fla is the momentum factor, lambda fla epsilon [0,1). Based on the gradient descent method, the weight coefficient update formula of the adaptive M-FLANN filter is as follows:

[0045]

[0046] make

[0047]

[0048] The weight coefficient update formula of the M-FLANN algorithm based on the momentum term technique is obtained by substitution as follows:

[0049] alpha p,i (n+1) = alpha p,i (n) - mu fla e fla (n)sin[p pi x(n-i)] + lambda fla [alpha p,i (n) - alpha p,i (n-1)]

[0050] beta p,i (n+1) = beta p,i (n) - mu fla e fla (n)cos[p pi x(n-i)] + lambda fla [beta p,i (n) - beta p,i (n-1)]

[0051] The fan pipeline noise reduction system of the double-loop active-passive coupling structure breaks through the bottleneck affecting the noise reduction effect, and promotes the wide application of ANC in pipeline noise reduction by reasonably designing the controller structure and optimizing the control algorithm.

[0052] The beneficial effects of the present application are:

[0053] First, a feedforward-feedback dual-loop control structure is constructed: the inner loop uses a parameter-autotuning notch filter array to eliminate mechanical harmonics, and the outer loop improves the FxLMS algorithm to suppress broadband aerodynamic noise, and dual-loop dynamic coupling control is achieved through an impedance matching network.

[0054] Second, the sinusoidal noise eliminator is used to effectively separate the narrowband noise component of the blade passing frequency and the broadband noise (turbulence and resonance components), thereby improving the stability of the system.

[0055] Third, by introducing momentum technology into the nonlinear ANC algorithm, an improved narrowband ANC algorithm is proposed. This algorithm not only speeds up convergence but also enhances the tracking performance of the system. Furthermore, key system parameters are optimized to ensure optimal noise reduction performance in various noise environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of a fan duct noise reduction system with a dual-ring active-passive coupling structure according to the present invention;

[0057] Figure 2 This is a diagram of the novel bilateral feedforward hybrid active noise control structure used in the outer loop of the present invention;

[0058] Figure 3 This is a structural flow chart of the active noise control method used in the inner loop of the present invention;

[0059] Figure 4 It is a schematic diagram of the overall structure of the present invention;

[0060] Figure 5 This is a noise reduction effect diagram of the present invention;

[0061] Figure 6 This is a waveform comparison diagram before and after noise reduction of the present invention. DETAILED DESCRIPTION

[0062] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0063] The flow chart of the fan duct noise reduction system with a dual-ring active-passive coupling structure in this specific embodiment is as follows: Figure 1 As shown, the following steps are included:

[0064] Step a: A double-layer duct structure is employed, with the inner duct wrapping the fan to initially reduce mechanical noise. Sensors are also placed on both sides of the fan to effectively control aerodynamic noise. The inner loop uses a parameter-autotuning notch filter array to eliminate mechanical harmonics, while the outer loop utilizes an improved FxLMS algorithm to suppress broadband aerodynamic noise.

[0065] Step b, for the aerodynamic noise, a new type of double-sided feedforward hybrid active noise control structure is adopted, an additional microphone is added on the intake side of the outer duct fan to capture another reference signal containing high-power blade passing frequency components and low-power turbulent noise, which greatly reduces the difficulty of extracting narrowband components. In addition, the input of SNC is a sinusoidal wave with a small amount of additive noise. The input and error signals are calculated as:

[0066] e f,k (n) = h k,0 (n) x f,k (n) + h k,1 (n) x f,k (n-1), k = 1, 2,... q

[0067]

[0068] where, is the MPA coefficient in SNC. Each MPA uses the LMS algorithm to update:

[0069] h k,0 (n+1) = h k,0 (n) + μ SNC,k x ω (n) x f,k (n)

[0070] h k,1 (n+1) = h k,1 (n) + μ SNC,k x ω (n) x f,k (n-1)

[0071] where, μ SNC,k is the step size used to update the filter coefficients. If the system weight update is accurate, the output e f,k (n) will converge to a single sinusoidal wave, which is the kth sinusoidal wave in the narrowband component x f,0 (n).

[0072] The error signal x ω (n) from SNC contains wideband components and is used as the input to the sub-controller W B (z). W B (z) is updated using the FXLMS algorithm:

[0073]

[0074] where, μ B is the step size for updating, and represent the length and filter coefficients, is S Z the impulse response coefficients of the filter of length M the estimate of the value of

[0075] As in the SNC, controlling each frequency component harmonic individually improves the prediction accuracy of its amplitude and phase, resulting in better stability and control performance. The coefficient output of each sub-controller is calculated as follows:

[0076] y f,k (n) = ω N,k,0 (n) x f,k (n) + ω N,k,1 (n) x f,k (n - 1), k = 1, 2,..., q

[0077] Unlike the coefficients in the SNC, each coefficient in the sub-controller is updated using the FXLMS algorithm:

[0078] ω N,k,0 (n + 1) = ω N,k,0 (n) + μ N,k e(n) x f,k (n)

[0079]

[0080] where μ N,k denotes another step size and e(n) denotes the residual error. Finally, the secondary source is used to cancel the loudspeaker emission to reduce the noise at the residual error microphone location. The auxiliary source y(n), y p (n), e(n), are calculated as follows:

[0081] y(n) = y ω (n) + y f (n)

[0082]

[0083] e(n) = p(n) - y p (n)

[0084] Step c, for motor noise, a momentum term factor is introduced in the FLANN system. At this time, the output signal of the FIR filter is:

[0085]

[0086] The FLANN performs nonlinear orthogonal expansion and weighted accumulation on the reference input signal through multiple linear combiners. Let the reference input signal be x(n), denote a first-order delay operation, represent the weight coefficient of the linear combiner, and the nonlinear orthogonal expansion of the p-th order for the input vector of length M is performed to obtain:

[0087]

[0088] where p represents the order of the non-linear extension term, p ∈ ([1, P1]), P1 is the maximum value of the order; [·] T denotes the transpose operation of a vector. In this case, the FLANN system output signal composed of P1 linear combiners is:

[0089]

[0090] The FLANN-based adaptive active noise control system output signal is composed of two parts, FIR output signal y fir (n) and FLANN processing signal y f (n), and the specific expression is:

[0091] y(n) = y f (n) + y fir (n)

[0092] The error signal is expressed as:

[0093] e fla (n) = p(n) - y fla (n)

[0094] The LMS algorithm is used to dynamically adjust the coefficients of the FLANN filter:

[0095] α p,i (n+1) = α p,i (n) - μ fla e fla (n) sin[pπx(n-i)]

[0096] β p,i (n+1) = β p,i (n) - μ fla e fla (n) cos[pπx(n-i)]

[0097] where μ fla is the update step size of the FLANN filter coefficient weight.

[0098] The cost function of the M-FLANN algorithm based on the momentum term technique is constructed as follows:

[0099]

[0100] where λ fla is the momentum factor, λ fla ∈ [0, 1). Based on the gradient descent method, the weight coefficient update formula of the adaptive M-FLANN filter is as follows:

[0101]

[0102] make

[0103]

[0104] The weight coefficient update formula of the M-FLANN algorithm based on the momentum term technology is obtained by substitution as follows:

[0105] alpha p,i (n+1)=alpha p,i (n)-mu fla e fla (n)sin[ppi x(n-i)]+lambda fla [alpha p,i (n)-alpha p,i (n-1)]

[0106] beta p,i (n+1)=beta p,i (n)-mu fla e fla (n)cos[ppi x(n-i)]+lambda fla [beta p,i (n)-beta p,i (n-1)]

[0107] The fan pipeline noise reduction system of the double-loop active-passive coupling structure breaks through the bottleneck affecting the noise reduction effect by reasonably designing the controller structure and optimizing the control algorithm, and promotes the wide application of ANC in pipeline noise reduction.

[0108] Figure 2 The novel double-sided feedforward hybrid active noise control structure diagram is adopted for the outer ring of the application.

[0109] Figure 3 The structure flow chart of the active noise control method is adopted for the inner ring of the application.

[0110] Figure 4 The overall structure schematic diagram of the application.

[0111] Figure 5 The noise reduction effect diagram of the application is shown in the figure. Figure 5 It can be seen that the application has good noise reduction effect.

[0112] Figure 6 The waveform comparison chart before and after noise reduction of the application is shown in the figure. Figure 6 It can be seen that the waveform after noise reduction of the application is obviously smaller than that before noise reduction.

[0113] The application provides a fan pipeline noise reduction system of a double-loop active-passive coupling structure, which has better noise reduction effect compared with the traditional pipeline fan noise reduction effect.

Claims

1. A fan duct noise reduction system with a double-ring active-passive coupling structure, characterized in that: The following steps are involved: Step a: A double-layer duct structure is employed, with the inner duct wrapping the fan to initially reduce mechanical noise. Sensors are also placed on both sides of the fan to effectively control aerodynamic noise. The inner loop uses a parameter-autotuning notch filter array to eliminate mechanical harmonics, while the outer loop utilizes an improved FxLMS algorithm to suppress broadband aerodynamic noise. Step b: For aerodynamic noise, a new bilateral feedforward hybrid active noise control structure is used. An additional microphone is added on the air inlet side of the outer duct fan to capture another reference signal containing high-power blade passing frequency components and low-power turbulence noise, which greatly reduces the difficulty of extracting narrowband components. In addition, the input of SNC is The input and error signals are calculated as: e f,k (n)=h k,0 (n)x f,k (n)+h k,1 (n)x f,k (n-1),k=1,2,...q in, is the MPA coefficient in SNC. Each MPA is updated using the LMS algorithm: h k,0 (n+1)=h k,0 (n)+µ SNC,k x ω (n)x f,k (n) h k,1 (n+1)=h k,1 (n)+µ SNC,k x ω (n)x f,k (n-1) Among them, μ SNC,k The step size used to update the filter coefficients. If the system weights are updated accurately, the output e f,k (n) will converge to a single sine wave which is the narrowband component x f,0 The kth sine wave in (n). The error signal x from the SNC ω (n) contains broadband components and is used as a sub-controller W B (z) input. W B (z) Update using the FXLMS algorithm: Among them, μ B is the step size used for updating, and express The length and filter coefficients are S Z The length is M, the impulse response coefficient estimated value. As in SNC, controlling each frequency component harmonic individually improves the prediction accuracy of its amplitude and phase, leading to better stability and control performance. The coefficient output of each sub-controller is calculated as follows: y f,k (n)=ω N,k,0 (n)x f,k (n)+ω N,k,1 (n)x f,k (n-1),k=1,2,…,q Different from the coefficients in SNC, each coefficient in the sub-controller is updated using the FXLMS algorithm: oh N,k,0 (n+1)=ω N,k,0 (n)+μ N,k e(n)x f,k (n) where μ N,k represents another step size, and e(n) represents the residual error. Finally, the speaker emission secondary source is eliminated to reduce the residual error noise at the microphone position. The auxiliary sources y(n), y p (n), e(n), are calculated as follows: y(n)=y ω (n)+y f (n) e(n)=p(n)-y p (n) Step c: In order to address the motor noise, a momentum term factor is introduced into the FLANN system. At this point, the output signal of the FIR filter is: FLANN performs nonlinear orthogonal expansion and weighted accumulation on the reference input signal through multiple linear combiners. Let the reference input signal be, represents the first-order delay operation, represents the weight coefficient of the linear combiner, and performs the p-th order nonlinear orthogonal expansion on the input vector of length, obtaining: Where p represents the order of the nonlinear expansion term, p∈([1,P1]), P1 is the maximum order; [·] T In this case, the output signal of the FLANN system composed of P1 linear combiners is: The output signal of the adaptive active noise control system based on FLANN is composed of the FIR output signal y fir (n) and FLANN processing signal y f (n) It consists of two parts, specifically expressed as: y(n)=y f (n)+y fir (n) The error signal is expressed as: e fla (n)=p(n)-y fla (n) The LMS algorithm is used to dynamically adjust the coefficients of the FLANN filter: a p,i (n+1)=a p,i (n)-m fla e fla (n)sin[pπx(ni)] b p,i (n+1)=β p,i (n)-m fla e fla (n)cos[pπx(ni)] Among them, μ fla is the update step size of the FLANN filter coefficient weights. The cost function of the M-FLANN algorithm based on momentum term technology is constructed as follows: Among them, λ fla is the momentum factor, λ fla ∈[0,1). Based on the gradient descent method, the weight coefficient update formula of the adaptive M-FLANN filter is as follows: make Substituting the weight coefficient update formula of the M-FLANN algorithm based on momentum term technology into the formula is: a p,i (n+1)=a p,i (n)-m fla e fla (n)sin[pπx(ni)]+λ fla [a p,i (n)-a p,i (n-1)] b p,i (n+1)=β p,i (n)-m fla e fla (n)cos[pπx(ni)]+λ fla [b p,i (n)-b p,i (n-1)] The proposed fan duct noise reduction system with a dual-ring active-passive coupling structure breaks through the bottleneck affecting the noise reduction effect by rationally designing the controller structure and optimizing the control algorithm, thereby promoting the widespread application of ANC in duct noise reduction.

Images