Solution Approach

In general, active noise and vibration cancellation works by injecting an equal but opposite signal into a system in order to counteract an undesirable noise and/or vibration. Therefore, an understanding of the magnitude and phase relationships in the system is critical in this endeavor. Although every situation is unique, the overall approach in implementing an ANVC system is similar from one application to the next. The basic approach includes the following high-level tasks:

• System Identification
• Actuator and Sensor Placement
• System Specifications Development
• Actuator Design
• Controller Algorithm Development and Implementation
• System Testing and Evaluation

Application Example

Cabin noise caused by the meshing of gear teeth inside the axle gear housing of some vehicles is a major concern for manufacturers. The gear teeth themselves are typically held to very tight tolerances already, so there is not much that could be changed with respect to the manufacturing of the gear housing. ETREMA worked with an axle manufacturer to develop an active vibration control system to reduce the front axle vibration on an all-wheel drive vehicle. The end goal of this work was to reduce the cabin noise resulting from this vibration.

The first task addressed in finding a solution to the gear mesh noise problem was system identification. This involved a modal analysis to correlate the input vibrations from the gear housing to the vibrations along the axle of the vehicle. This analysis provided transfer function information so that the output at a given location could be predicted from a known input. The data was collected via the SigLab data acquisition system and analyzed using STAR modal software, both from Spectral Dynamics, Inc.

The propagation path of the noise was identified next in order to correlate the noise in the cabin of the vehicle with the vibrations along the axle. For this task, ETREMA engineers used Matlab, a software package from The Mathworks, Inc., to analyze acceleration and acoustic data collected with the SigLab system. This information was used to determine the most effective placement of actuators and sensors along the axle in order to affect the greatest reduction in gear mesh noise as heard in the vehicle cabin.

The requirements generated in the previous step dictated the specifications of the actuator to be used to counteract the undesirable vibrations in the axle. In some cases, one or more of ETREMA's standard actuators may be used in the application if the specifications already meet the requirements. Otherwise, a custom actuator may need to be developed. In particular, if there are unique environmental issues that must be addressed, some modifications may be required.

The next step in solving a noise and/or vibration problem with active means is to identify a control algorithm that suits the requirements of the system. There are many algorithms to choose from and many are well documented in the industry literature. In the vehicle application, an adaptive feed-forward algorithm that utilized the shaft rotation as a reference signal was incorporated. A Least Mean Square (LMS) algorithm was also used to minimize an "error" signal, where the error signal was the acceleration at a given point along the axle that had a strong correlation to the cabin noise. A diagram of the algorithm is shown below where the solid red blocks and lines represent actual hardware and the blocks and lines in dashed blue represent calculations made by the controller algorithm software. The "Estimate of the Controller Plant Transfer Fcn" in the figure is an estimate of the actual transfer function between each actuator and each sensor, calculated from measured data. Variable definitions are:

r(k) Reference signal Ref. signal filtered by an estimate of the control plant u(k) Control signals sent to actuators (control plant input) Vector of filter weights (2 weights per frequency) yr(k) Existing gear mesh vibration (disturbance plant output) yu(k) Vibration due to control actuators (control plant output) y(k) Measured output (total vibration signal) d(k) Desired system output (in this case, d(k) = 0) ε(k) Error signal

Diagram of the algorithm used in the vehicle ANVC application.

1. The reference signal, r(k), is filtered by an estimate of the transfer function between each actuator and sensor, resulting in .
2. Through the use of a Least Mean Square (LMS) algorithm, this filtered reference signal, , is used in conjunction with the error signal, ε(k), to update the filter weights, , of a Finite Impulse Response (FIR) filter.
3. The control signal is calculated by passing the reference signal through this FIR filter.

The final step in the ANVC solution is the implementation of the system. This includes subsystem testing as well as full integration Small adjustments to the controller coefficients are usually necessary to account for unforeseen interactions between system components and any differences between the theoretical design and the actual hardware built.

In the vehicle application, at approximately 2500 rpm (corresponding to 53 mph), the Sound Pressure Level (SPL) at the driver's right ear dropped by 6 dBA, from 64 to 58. For typical human hearing, a drop of 6 dBA is perceived as half as loud. At other speeds, the SPL was either less than the baseline run (controller turned off) or was below the 50 dBA noise floor of the vehicle.