طراحی استراتژی جدید برای یادگیری تکرار ی و کنترل کننده های تکراری سیستم های با چگالی معین بالا: استفاده برای کنترل سر و صدا فعال

This paper describes the application of a novel design strategy for iterative learning and repetitive controllers for systems with a high modal density, presented in the companion paper, on two experimental case studies. Both case studies are examples of active structural acoustic control, where the goal is to reduce the radiated noise using structural actuators. In the first case study, ILC is used to control punching noise. An electrodynamic actuator on the frame of the punching machine is driven by the ILC algorithm which takes advantage of the repetitiveness of the consecutive impacts to reduce noise radiation. In the second case study, an RC algorithm is used to control the noise radiated by rotating machinery, which is often mainly periodic. A piezoelectric actuator incorporated in the bearing is driven by the RC algorithm which is capable of reducing harmonics of the rotational frequency of the shaft. Both applications show the practical usefulness of the novel design strategy. Keywords

Noise pollution from modern industrial activities is an environmental problem of growing importance. At the moment, these noise problems are mostly addressed in a passive way (e.g. encapsulations, absorbing material, etc.). However, passive noise control techniques are inefficient at low frequencies, resulting in large installation spaces. Therefore, in certain applications, active solutions are a potential alternative, suitable for the reduction of low frequency noise. In contrast to passive control methods, which only use passive elements and consequently do not add energy to the controlled system, active control techniques rely on an external energy source to reduce the radiated noise. Depending on the type of actuators (acoustic/structural), two control strategies can be distinguished for the active control of noise: active noise control (ANC) and active structural acoustic control (ASAC). In ANC [1], acoustic actuators such as loudspeakers are used to generate a secondary sound field (the anti-noise) with the aim of attenuating a primary, disturbing sound field. ASAC, where structural actuators intervene in the vibrational pattern of a structure with the objective of attenuating the radiated noise level, can be an alternative, more efficient approach to tackle structure-borne noise problems [2]. Generally less actuators are required to achieve structure-borne noise reduction if ASAC is applied instead of ANC. Two application fields where the potential of ANC and ASAC is evaluated nowadays, is the control of periodic noise radiated by rotating machinery [3] and the control of transient noise [4]. The most popular control algorithm in ANC and ASAC systems for rotating machinery is the filtered-X least mean squares (FxLMS) algorithm, which is an adaptive feedforward control technique. An adaptive technique is required since the controlled acoustic systems are often subject to changes: the spectral density of the disturbance is nonstationary, the controlled plant can vary substantially due to environmental influences, etc. The algorithm is based on the availability of a reference signal, like the fundamental frequency of the disturbing noise. In applications where the disturbance is periodic (e.g. rotating machinery) this can usually be provided by a tacho signal. This reference is filtered through a control filter to calculate the driving signal for the secondary actuator. In order to cope with the changes in the controlled system, the control configuration is augmented with additional sensors, installed at the locations where noise reduction is required. Based on these additional sensor signals, the control filter is adjusted to the varying primary noise field. The potential of the FxLMS-algorithm to obtain periodic noise reduction has been demonstrated in many applications [5]: active duct outlets [6], active control systems to suppress engine noise in car interiors [7], active control systems for the reduction of the propeller noise in aircrafts [8], etc. However, the performance of the FxLMS significantly decreases when the disturbance is transient instead of stationary. Adaptive algorithms require a certain adaptation time before convergence and optimal performance is attained. This adaptation time is not available in the case of transient noise, which explains the unsuitability of the popular FxLMS-algorithm for transient noise control purposes. In [9], it is even stated that the FxLMS-algorithm may become unstable in cases where the primary noise is impulsive. As a result, new types of algorithms are necessary, which are adapted to the specific transient character of the disturbance noise. Especially in machine halls with production machines that generate impact noise, the radiated noise levels often exceed the ever more restrictive legal regulations regarding human exposure to noise. Examples of production machines radiating excessive impact noise levels are punching machines, forming presses, bending machines, drop forges, etc. In many of these impact noise problems, the same mechanical operation generates successive radiated noise pulses, which consequently have a repetitive character. Since iterative learning control (ILC) is a well-known technique to suppress a repetitive disturbance acting on a system [10], the potential of this technique with respect to the control of repetitive transient noise will be investigated in the first part of this paper. In the second part of the paper, RC, which is a control technique closely related to ILC suitable for suppressing periodic disturbances [11], is presented as a valuable alternative for the traditional FxLMS algorithm in the control of rotating machinery. The applicability of the traditional ILC and RC design approaches, like the inverse model-based approach [12], the optimization-based approach [13] and [14] and the frequency-domain approach [15], is limited for the control of acoustic systems, which are typically characterized by a high modal density with lowly damped complex poles and zeros, possibly exhibiting a time delay. All these design approaches rely on a parametric system model, which is difficult to estimate in the case of acoustic systems. To bridge this gap, in the companion paper [16], a novel design strategy is presented, specifically adapted to such systems, and extended to RC design. In the procedure, the main objective is to fulfill the convergence criterion in a selected frequency range. This results in a controller which is robust for plant uncertainty and most effective at the system's resonance frequencies, a useful property in noise control applications. The goal of this paper is to experimentally assess the performance and robustness of this novel strategy in ASAC applications. It will be shown that the controllers are robust for plant uncertainty and the influence of an inaccurate trigger signal in the ILC algorithm will be investigated. Therefore, this paper presents the application of this novel strategy in two case studies. In the first application, ILC is applied to control the transient noise of punching machines. The consecutive impacts are repetitive and the effect of one impact has died out before the next impact occurs, such that they can be considered as separate phenomena. With both requirements fulfilled, ILC is a suitable strategy for this application. In the second application, RC is applied to control a piezoelectric bearing in order to reduce the periodic noise radiated by rotating machinery, where the frequency of the noise is determined by the rotational speed.

This paper presents two ASAC applications, iterative learning control of punching machines and repetitive control of rotating machinery, in which the controllers are designed according to the methodology proposed in the companion paper [16]; a novel frequency-domain ILC and RC design methodology, adapted to systems with a high modal density, typically considered in active structural acoustic control (ASAC) and active noise control (ANC) applications. Since the impacts generated by a punch are repetitive and separate phenomena, ILC is a suitable control strategy. With the novel design strategy, noise level reductions up to 4 dB can be achieved. Theoretically, more reduction can be achieved, but the limiting factor is the accuracy of the trigger signal which announces the next punch. In the second application, an active bearing is used to reduce noise generated by rotating machinery. With the RC design strategy, reductions in the vibration level up to 15 dB can be achieved, resulting in significant reductions in the noise level. Both applications show the applicability of the design procedure for systems with a high modal density.