Road quality and ride comfort are major concerns when creating and maintaining roads. Ride comfort is dependent on the interaction between the vehicle and the roughness of the road. Road roughness is currently measured by road-profiling vehicles in a quantifiable term known as the International Roughness Index (IRI). Although this method is useful for determining road surface information, it is a time consuming process, it can’t be carried out every day, and it does not provide a direct indication of ride comfort.
However, advancements in sensor technology provide necessary enhancements that current methods cannot address. This study aims to develop an innovative method using built-in wireless sensing and mobile computing features of smartphones to not only estimate road roughness, but to provide a direct real-time indication of ride comfort. Estimation of road roughness based on vehicle response involves insight regarding the properties of the vehicle itself.
While the vibration response of the vehicle can be readily measured using wireless accelerometers or built-in smartphone sensors, information pertaining to the vehicle and road properties is left unknown. To address this issue, various system identification methods are evaluated for high-damped systems and applied to the vehicle. Through the application of system identification methods using vehicle response data, the unknown parameters of the vehicle can be estimated.
These methods are validated through analysis of vehicle model simulation paired with standard simulated road profiles. Furthermore, these simulations create an environment to determine optimal conditions for vehicle mass prediction. With vehicle parameters identified, the dynamic response parameters of the vehicle and the input of the road surface profile can be correlated to estimate the IRI while directly providing ride comfort information. Field testing involving the use of a wireless accelerometer and GPS is also implemented to compare recorded data against the simulation findings.
This study establishes a framework that integrates wireless sensors, system identification methods, and the correlation between ride comfort and the IRI with vehicle vibration measurements. System identification methods with a focus on vehicles subjected to excitation from the road are evaluated. This involves an investigation of prediction error identification methods with the use of grey-box modeling to estimate vehicle mass under varying road conditions.
With vehicle parameters known, correlation of vehicle vibration response with the IRI and ride comfort is empirically established to determine areas of road in need of maintenance along with comfortable travel routes in real-time. This study demonstrates the appeal for including information related to road conditions and ride comfort in mobile maps for alternative travel routes.
VEHICLE INPUT / OUTPUT
Before the parameters of a vehicle can be analyzed, an understanding of the various vehicle inputs and outputs must be understood. As seen in Figure 2.1, the vehicle can be viewed as a system. This system has both an input and an output. In the case of the vehicle, the input is the changing road profile elevation, while the output is viewed as the vibrational response.
As visualized in Figure 2.3, this profiler makes use of an accelerometer in conjunction with devices such as laser transducers, and infrared and ultrasonic sensors to measure the height of the ground beneath the vehicle. The accelerometer is used to measure the vertical acceleration as an inertial reference point for the relative height to the ground.
The human body can be viewed as a linear passive mechanical system for small amplitude vibrations. Vibrations up to 12 Hz affect all of the human organs, while local effects occur at higher frequencies (Hostens, et al., 2003). For subjects exposed to vibrations in the vertical direction, resonance changes depending on body position. For a seated person, the first resonance occurs between 4 and 6 Hz, while for a standing person, resonance occurs at both 6 and 12 Hz (Piersol, et al., 2009).
Through the use of an impulse force, the Eigensystem Realization Algorithm (ERA) excludes the input and feedthrough matrices in the state-space model for modal identification. The Natural Excitation technique is used with ERA to simulate this impulse response due to excitation from a white noise input load (Chang, et al., 2012). This method has been found to be accurate, but requires impulse for free response data under noiseless conditions (Petsounis, et al., 2001). Additional steps and data may need to be taken to account for the distortion due to noise.
Numerical algorithms for subspace state space system identification, known as N4SID, compare projections of future output against recorded output data (Van Overschee, et al., 1993). The predicted state-space model is found once the projection of the future output and recorded output has been minimized. This process of parameter identification through eigenvalue estimation is shown in in Figure 3.2 (Chang, et al., 2012).
Other popular linear models include the AutoRegressive Moving Average (ARMA) model and AutoRegressive Moving Average with eXogenous excitation (ARMAX) model. These methods use past values in the time series to predict future behavior of the response.
As of 1995, of the 600,000 bridges located in the United States, 150,000 of them had bumps formed at the ends (Briaud, et al., 1997). Simply put, these bumps at the ends of the bridges are caused by a differential in settlement at the joint connecting the bridge approach and the bridge deck, as shown in Figure 4.28. Major causes of the bump include compression of the fill material, settlement of the soil beneath the embankment, design and construction errors, high traffic loads, and poor drainage.
Additional causes of bump formation are shown in Figure 4.29. Some sources state that at a height differential of 12.7 mm (0.5 in.) between the bridge deck and road pavement, the bridge requires maintenance (Dupont, et al., 2002). A 1995 survey conducted in Illinois showed that more than 27% of bridges had bumps larger than 50.8 mm (2 in.) (Stark, et al., 1995).
Further simulation is completed to express the range of discomfort based on both a change in bump height and vehicle speed. These points represent the peak frequency weighted acceleration for a vehicle traveling along a road with no roughness aside from the step representing a bridge bump. This is done to isolate the effect of the bump without any additional effects from other sources of roughness. The results displayed in Figure 4.36 are split to show varying ranges of comfort.
Determining the effectiveness of system identification for unknown vehicle mass is accomplished by collecting the vertical acceleration response through wireless sensing and processing the data with PEM analysis. Testing is performed on a variety of surfaces for this study to understand the behavior of the vehicle based on the input. This includes road surfaces ranging from smooth to rough and damaged pavements. In addition to these surfaces, known inputs are also used, such as speed bumps, bumps at the ends of bridges, and an extension cable used to simulate a pneumatic road tube.
All testing in this study is performed using a 2009 Toyota RAV4. For data collection, the PASCO PS-2119 acceleration sensor is firmly attached to the vehicle’s center console using Velcro straps as seen in Figure 5.11. In addition to the use of straps, the sensor is pushed down by hand to prevent additional vibration. The sensor is attached to the center console rather than one of the seats due to consistency of data results.
Performing this test is accomplished through the use of a 6.35 mm (0.25 inch) diameter extension cable in place of a standard pneumatic road tube as shown in Figure 5.27. The cable is firmly stretched and taped in the center and on each end to prevent movement while the vehicle passes. This test is performed in a parking lot to prevent additional road roughness from interfering with the input of the cable.
Once the vehicle parameters are known, either through system identification or previously known properties, the relation of the acceleration response can be directly related to the roughness of the profile. The relation between the vehicle output response and the IRI is shown in Figure 6.3.
A color scale is created to visualize the IRI and comfort levels on the profile map. Two individual scales are established that relate to the IRI roughness descriptions and the comfort described by the frequency weighted accelerations. The selection of IRI values is based on an approximate IRI for various road classes shown in Figure 2.20.
High-speed road profilers are currently used to measure the elevation of road surfaces to calculate road roughness, but do not provide a direct measure of ride comfort of the user. Due to technological advancements, mobile sensing has become a viable tool for measuring vibrations, which can be applied to the analysis of road conditions and ride comfort. It addition to providing another layer of information by including ride comfort, key road information can be updated in real-time through the use of mobile sensing, whereas current profiles are updated between every one to two years.
The research completed in this dissertation demonstrates the viability of estimating road roughness and directly measuring ride comfort through wireless sensing. Due to the variation in vehicle properties, this study focuses on parameter identification of high-damped systems. This includes the evaluation of system identification methods and the necessary requirements to obtain an accurate prediction of the vehicle mass by applying these methods.
Recommendations for Future Work
- This study presents the necessity of accurately estimating the input of the vehicle when applying system identification methods for prediction of vehicle parameters. Further research can be conducted to estimate the profile input based additional information, such as natural frequencies and damping ratio of the vehicle found in the output acceleration response.
- Although the half car model provides a worthwhile representation of the vehicle for parameter identification studies, it can be further extended to a full car model to account for lateral effects due to road unevenness and mass position of passengers and cargo.
- For this study, the sensor is assumed to be located at the center of gravity of the vehicle to negate effects of vertical vibration due to vehicle pitch (rotation). A further look into the influence of sensor location can demonstrate the effect of vehicle pitch on the recorded response.
Source: University of Maryland
Author: Yunfeng Zhang