Application of Vehicle Mounted Accelerometers to Measure Pavement Roughness

Abstract

To enhance the efficiency of pavement roughness measurement and reduce the cost, an integrated and wireless transfer based measuring system was developed. The proposed system can obtain vehicles status and location data via wireless acceleration sensors and GPS, calculate the international roughness index (IRI) by power spectral density analysis, and provide reports automatically. This paper presents the architecture of the proposed system, consisting of data collector, car mounted terminal, and information platform. Two wireless communication systems (ZigBee and 3G modules) were utilized to transfer the data and construct network between the components. The information platform implemented an acceleration-IRI model to calculate IRI, and a GPS based distance algorithm was employed to segment the measured road per 1 km. The various results are saved in an Oracle database, displayed on the digital map and made available to the mobile terminal. Several field tests of the prototype system were conducted in Huzhou, Zhejiang province in China. The results show that, compared to the laser roughness testing method, the relative error of this proposed system is less than 10%, which verifies the accuracy, effectiveness, and reliability of the proposed measuring system.

1. Introduction

Pavement roughness significantly affects riding quality, driving safety, and fuel consumption. Highway authorities generally concentrate on reducing pavement roughness to improve driving quality and minimize related losses. However, without suitable evaluation of pavement roughness, road maintenance and rehabilitation are passive and aimless.

The international roughness index (IRI) was proposed by Sayers et al. [1] to measure road roughness. It is a well-recognized standard scale widely used in many countries. Several practical techniques have been proposed and implemented, which fall into three types: direct profile, indirect profile, and response type measurements [2], ranging from simple rod and level static measurements to more advanced high-speed noncontact surveys [3]. The relevant devices are difficult and/or expensive to operate and maintain, and the test cycles are relatively long and not sufficiently efficient to provide useful feedback to authorities. Therefore, we propose a new measurement method incorporating accelerometers combined with wireless communication and GPS to measure pavement roughness rapidly and efficiently.

The first attempts to measure pavement roughness, in the early years of the last century, were by a sliding straightedge [4]. Many mechanical devices were subsequently developed to improve measurement accuracy and efficiency. Direct profile measurements generally evaluate pavement roughness by directly measuring changes of elevation from the road surface, such as rod and level profile and longitudinal profiler (TRRL Beam). These measurements can provide precise IRI and can be used to calibrate other devices. However, they have obvious disadvantages for field tests and are inadequate for a wide range of measurements. Indirect profile measurements were designed to measure longitudinal profile over the wavelength range of interest [5]. They largely quantize displacement of vehicles caused by pavement roughness, such as the General Motors Research (GMR) profilometer and longitudinal profile analyzer (APL). These devices are more efficient than direct profile measurement but remain sensitive to the measurement environment. Response type measures pavement roughness by correlating the vehicles response (displacement, acceleration, etc.) caused by roughness to the IRI calculated from a profile, such as Bump Integrator [6–8]. These devices can effectively improve the measurement efficiency, but the time stability is relatively poor and easily gets affected by environment [9]. In this paper, we proposed a power spectral density (PSD) method to calculate IRI using acceleration data [10, 11], which is more accurate and stable than the traditional integral method.

Wireless sensor networks have been widely used in civil, transportation, and environmental engineering for structural and environmental monitoring and internet connection for vehicles [12]. ZigBee wireless transmission has experienced rapid development due to its high reliability and low power consumption, making it suitable for intervehicle communication [13, 14]. However, 3G wireless networks provide higher transmission bandwidths and longer transmission distances, which are appropriate for vehicle-server communication.

Recent years witnessed the rapid progress in dedicated short-range communication and wireless sensors, which laid cornerstones for vehicular cyber-physical systems (VCS) [15]. VCS can be regarded as an advanced perception and communication method for vehicles by using intelligent and interactive operations [16]. With the technique of cloud computing, vehicular data can be delivered and analyzed in real time, impelling various data-based applications [17].

The components of the measuring system are introduced in Section 2, and the design details of the data collector are provided in Section 3. The vehicle mounted wireless sensor network based on ZigBee devices is introduced in Section 4. Data processing and publishing are detailed in Section 5. The field test of the proposed system is described and discussed in Section 6, and we present our conclusions in Section 7.

2. Measuring System Components

An integrated, wireless transmission based system was developed to measure pavement roughness. The proposed system is shown schematically in Figure 1 and incorporates three components: data collector, car mounted terminal, and processing.

Figure 1

Schematic diagram of the measuring system.

z-axis accelerations are collected then transferred to the car mounted terminal via the ZigBee module. To improve data accuracy and comprehensively take different wheel track tapes into consideration, two data collectors were fixed right above the left and right rear wheels.

The location and vehicle status information are obtained by gyroscope, angular acceleration sensor, and GPS within the car mounted terminal. These data are combined with the accelerometer data, preanalyzed and packaged within the terminal, and the packet is sent to the FTP server by the embedded 3G/4G module.

Within the server, the processing system analyzes and evaluates the data, finally producing the IRI. The calculated IRI can be displayed on a digital map at the server and also provided to road users, which plays an important role in route choices of ambulance and the cars with babies.

3. Data Collector (Wireless Sensor)

3.1. Working Principle

The data collector comprises three sections: z-axis accelerometers (type MMA8451Q), microprocessor control unit (MCU) (type TC12C5608AD), and ZigBee module, as shown schematically in Figure 2, and the workflow is shown in Figure 3.

Figure 2

Schematic diagram of data collector.

Figure 3

Workflow of the data collector.

The accelerometer is a triaxial acceleration sensor which can obtain x, y, and z-axis accelerations. The accelerometer data speed may be between 1 and 100 Hz, since the driving speed was about 50 km/h, which equals 15.3 m/s. If the frequency was set to 20 Hz, the acceleration data would be obtained about every 0.76 m, which is accurate enough for measuring. And considering the stability of data transmission, lower frequency can reduce the possibility of data loss; thus the frequency employed was 20 Hz.

The MCU connects the accelerometer and ZigBee module. As accelerometer data is collected, the MCU analyzes and packages the data and then transmits the packages through the ZigBee module. The MCU can also receive control instructions and determine transmission frequency and status of the ZigBee module.

The ZigBee module is used to send and receive data or instructions. ZigBee transfer offers more convenience than wire transmission-way. The low power requirement and short distance are suitable for in-car communication.

3.2. Communication Protocols

The communication protocols define data transmission between accelerometers, MCU, and the ZigBee module. The ZigBee module sends control instructions to the MCU, as shown in Table 2, where the 0x00~0xAA are the six instructions (hexadecimal), and the MCU sends acceleration data, detailed in Table 1, to the ZigBee module at a different frequency; the size of acceleration data is 5 bytes which contains frame header (0xAA), sensor number (F0–F7), x-axis acceleration data (2 bytes), and z-axis acceleration data (2 bytes). The measuring range of the accelerometers is 2 g (>256); thus it costs 2 bytes to save the acceleration data which contains high-order byte (X_MSB, Z_MSB) and low-order byte (X_LSB, Z_LSB). Table 1(b) shows the structure of the acceleration data. The data are based on eight-bit byte; in Table 1(b), “0” and “1” are the fixed bit, while “x” is the variable bit.

Table 1

Data package.

(a)
Byte 0		Byte 1		Byte 2		Byte 3		Byte 4	Byte 5
Start		Sensor number		x-axis (raw)			z-axis (raw)
(b)
	Bit 7	Bit 6	Bit 5	Bit 4	Bit 3	Bit 2	Bit 1	Bit 0	Comment
Byte 0	1	0	1	0	1	0	1	0	0xAA
Byte 1	1	1	1	1	0	x	x	x	F0–F7
Byte 2	0	0	x	x	x	x	x	x	X_MSB
Byte 3	x	x	x	x	x	x	x	x	X_LSB
Byte 4	0	0	x	x	x	x	x	x	Z_MSB
Byte 5	x	x	x	x	x	x	x	x	Z_LSB

Table 2

Control instructions.

Control	Function	Remarks
0x00	Reset	Restore the default state
0x01	Set the time interval to 50 ms	Set the frequency of data input to 20 Hz
0x02	Set the time interval to 100 ms	Set the frequency of data input to 10 Hz
0x0A	Turn on noise-reduction mode	Default
0x0B	Turn off noise-reduction mode
0xAA	Request data sending

3.3. Component Locations

Acceleration data are different at different places within the vehicle. However, since the vehicle is a rigid structure, if a large acceleration is produced at one point, corresponding accelerations will inevitably appear at other points. Thus the single-point measuring method is not accurate enough to represent the pavement roughness of whole road cross section. Therefore, the multipoint method was proposed to correlate the influences.

The full-size vehicle can be simplified to linked four-quarter-car model, the corresponding parts of the vehicle rigid body above the each wheel are the unsprung masses of the four-quarter-car model, and the acceleration of the four parts is the systematic responses of the four-quarter-car model. Therefore, considering the vehicle structure, the influence of engine vibration which occurs during measurement, and many field tests, we installed two accelerometers on the rigid vehicle body and above the left and right rear wheels, and, in consideration of the stability of data transmission, the two accelerometers were fixed on the inside of the vehicle, as shown in Figure 4. Since most measuring vehicles are front wheel drive, proposed accelerometer location minimizes the influence of engine vibration.

Figure 4

Location of the accelerometers.

4. Car Mounted Terminal

4.1. Working Principle

The car mounted terminal acts as a relay station in the data transmission, and several modules and sensors are integrated:

• (1)

GPS module.

(2)

Angular velocity sensor.

(3)

Geomagnetic sensor.

(4)

ZigBee module.

(5)

Acorn RISC Machine (ARM) development board (S5PV210A ARM A8, 1 GHz).

(6)

3G module.

(7)

Touch screen.

The GPS module obtains the location and time. The angular velocity and geomagnetic sensors were used to calibrate the GPS data. Positional accuracy was approximately 1 m, which is sufficient for field measurement.

Accelerometer data packages are received by the ZigBee module, and acceleration, location, and so forth can be displayed on the screen while the multiple data is analyzed and packaged by the ARM development board. The operator may record, save, and send the data package using the touch screen. The 3G module was used to upload the data package to the FTP server and for subsequent forward to the software system for analysis.

The car mounted terminal includes four processes, which are executed serially, initialization, recording, saving, and exit, as shown in Figure 5.

Figure 5

Car mounted terminal processes.

4.2. Data Matching

The multiple sensor data must be matched and packaged. This process operates in parallel with GPS and acceleration data recording and we record the data to txt-(ASCII) files.

The raw data obtained includes three-axis acceleration data. z-axis acceleration shows the vertical vibration of the rigid vertical body, x-axis acceleration shows the driving condition of the vehicle, and y-axis acceleration shows the lateral movement of the vehicle. The vertical vibration reflects the roughness of the pavement, and the driving condition of the vehicle determines the driving trail. Thus the x-axis and z-axis accelerations were required for analysis. To facilitate the calculation and locating, the corresponding GPS data and time and speed data were packaged.

During the measuring, at first, a data queue is created; the GPS data is captured and written into head of the queue. Then the terminal starts to receive acceleration data stored in this queue until next GPS is captured. In this way, we can guarantee that all received acceleration belongs to the interval of these two GPS coordinates. While all the data is captured and analyzed, the queue is going to be written into a txt-file. The time interval of the recording process is 1 s. Therefore, the data are matched and written each second. The format of the matched data is shown in Table 3.

Table 3

Matched data format.

Line	Start	Contents	Separator
1	#	Time, longitude, latitude, speed	,
2	$\ast \mathit{1}X$ :	x-axis acceleration array of sensor 1	,
3	$\ast \mathit{1}Z$ :	z-axis acceleration array of sensor 1	,
4	$\ast \mathit{2}X$ :	x-axis acceleration array of sensor 2	,
5	$\ast \mathit{2}Z$ :	z-axis acceleration array of sensor 2	,

5. Data Analyzing and Publishing

5.1. Server Construction and Operating Process

Once the data are matched and saved, the car mounted terminal sends a connection request to the server before instigating data transmission. When the request is received, the terminal must be authenticated. Once the data is received and corresponding files are created on the server, the data may be analyzed, saved, and displayed.

The server comprises three modules, as shown schematically in Figure 6:

• (1)

Listening module: it listens on the port.

(2)

Recording module: it receives and records the uploaded data.

(3)

Monitoring module: it monitors the running state of the system.

Figure 6

Flow chart of server modules.

The server comprises four processes: environment initialization, thread initialization, authentication, and data transmission, which are executed serially. Environment initialization resolves the IP address and creates the TCP listening thread; then thread initialization creates the TCP client and waits for the link. Authentication must be executed before linking; then data transmission commences and the uploaded data is saved on the server. Finally, the developed software calls the file and calculates pavement roughness using the proposed algorithm (Section 5.3). The web based platform publishes the roughness status, and the results may be shown visually in digital maps.

5.2. Pavement Roughness Algorithm

We adopted IRI to measure the pavement roughness, calculated using the power spectral density (PSD) method. We must consider the effects of engine vibration, accelerometer location, driving speed, and system error within our algorithms to ensure measurement accuracy.

IRI was established by the World Bank in 1986 and is calculated using a quarter-car model, where relative displacement between the sprung and unsprung masses during driving is accumulated to yield the roughness index with units of slope (m/km). The index reflects the accumulation of vehicle vibration amplitude caused by the pavement as

$$\begin{array}{cc}& \mathrm{I}\mathrm{R}\mathrm{I}=\frac{\mathrm{1}}{L}{{\displaystyle \int }}_{\mathrm{0}}^L\left|Z_s-Z_u\right|dl,\end{array}$$

(1)

where L is the measured distance and $Z_s$ and $Z_u$ are the displacements of the sprung and unsprung masses, respectively.

The acceleration-IRI model [8] considers the approximate linear relation between the mean squared value and IRI and the correlation between the displacement and acceleration spectra ( $S_a\left(\omega \right))$ as

$$\begin{array}{cc}& \mathrm{I}\mathrm{R}\mathrm{I}=A·{\phi }_Z=A·\sqrt{{{\displaystyle \int }}_{-\infty }^{+\infty }{\left|H_{ZY}\left(\omega \right)\right|}^{\mathrm{2}}{\left(\frac{v}{\omega }\right)}^{\mathrm{4}}{S}_a\left(\omega \right)d\omega },\end{array}$$

(2)

where v is measuring speed, ω is angular frequency, ${\phi }_Z$ is the mean square value of acceleration spectra, $H_{ZY}\left(\omega \right)$ is the frequency response function of $Z(t)$ , and $Z(t)=\mathrm{\hspace{0.17em}\hspace{0.17em}}{Z}_s(t)-Z_u(t)$ . When the driving speed is constant, IRI is approximately linearly related to the mean squared acceleration spectrum. The linear coefficient can be determined by field tests.

As discussed above, we adopted a two-wheel location for the data collectors and accordingly propose a two-wheel correction for the model. We used multiple linear regression (MLR) method to fit IRI to the two-accelerometer data as

$$\begin{array}{cc}& \mathrm{I}\mathrm{R}\mathrm{I}=\mathrm{5.062}{X}_l+\mathrm{6.471}{X}_r-\mathrm{0.1651},\end{array}$$

(3)

where $X_l$ and $X_r$ are the square root of the left and right wheel acceleration spectral densities, respectively. R-square value is 0.937, which is reasonable for IRI measuring.

It is not practically feasible to keep constant speed (80 km/h) as required in the IRI definition. Thus, we propose a velocity correction. Due to the complexity of the speed-influence, maximum likelihood was used to correct for speed effects. After several field tests, we propose a velocity correction coefficient as

$$\begin{array}{cc}& \alpha =\frac{\mathrm{1}}{\sqrt{\mathrm{0.0003865}{v}^{\mathrm{2}}+\mathrm{0.0009125}v}},\end{array}$$

(4)

where v is the actual instantaneous driving speed (km/h). The IRI model becomes

$$\begin{array}{cc}& \mathrm{I}\mathrm{R}\mathrm{I}=\mathrm{5.062}\alpha X_l+\mathrm{6.471}\alpha X_r-\mathrm{0.1651}.\end{array}$$

(5)

Although the velocity correction model can modify the effect of speed, the suggested driving speed range is 55–60 km/h.

5.3. Software Development

The software system developed in-house, shown in Figure 7, was used to calculate IRI, link the database, and show the results in real time. It comprises five components, as shown in Figure 7:

• (1)

Data acquisition.

(2)

Calculation.

(3)

Database.

(4)

Digital map display.

(5)

Automatic report generation.

Figure 7

(a) Roughness measurement. (b) Roughness examination and repair.

The data acquisition module monitors the data folder on the FTP server. Once the data file is obtained, the module reads and analyzes the file automatically. The calculation module, which includes the calculation algorithm described above, analyzes the uploaded data and generates maintenance suggestions based on the results. We calculate IRI per km over the measured range and define the measured section by GPS data per kilometer. Speed information reflects the vehicle driving status. When the speed is less than 5 km/h, we consider the vehicle to be stationary and the corresponding data is filtered out. The IRI of each section is calculated via the algorithm described above. The core algorithm includes filtering, Fourier transform, and PSD algorithm and was written in Matlab and then invoked by the main program and into C#.

The calculation results are saved in the Oracle database, including the raw data and corresponding repair suggestions. The digital map module displays the results, including IRI and measuring tack. The module was compiled in C# and JavaScript, and Baidu Map API was invoked to transform the coordinates, locations, and display. The measurement report is automatically generated by the final module as showed in Figure 8.

Figure 8

Flow chart of the software system.

6. Field Test

To verify the measurement accuracy and test the system operation, twenty sections of road arteries in Huzhou were chosen for the field test. The data for the sections were provided by the Huzhou Highway Administration Bureau. The Huzhou Highway Administration Bureau adopted the RQI (Riding Quality Index) to represent pavement roughness, so we transformed the IRI to RQI, since

$$\begin{array}{cc}& \mathrm{R}\mathrm{Q}\mathrm{I}=\frac{\mathrm{100}}{\mathrm{1}+\mathrm{0.0185}{e}^{\mathrm{0.58}·\mathrm{I}\mathrm{R}\mathrm{I}}}.\end{array}$$

(6)

Then we compared the transformed RQI and the actual RQI (measured by road laser profilometer; Highway Administration of Huzhou, 2014) to test the accuracy of the system, as shown in Table 4. The proposed measurement system shows good agreement with RQI (measured by laser roughness testing vehicle), with largest relative error less than 10%.

Table 4

Proposed system and actual RQI.

Number	Measured RQI	Actual RQI	Relative error (%)
1	93.78	92.6	1.18%
2	92.41	94	1.59%
3	92.49	95	2.51%
4	92.25	94.7	2.45%
5	94.98	94.7	0.28%
6	94.67	95.4	0.73%
7	94.15	95.5	1.35%
8	91.52	94.2	2.68%
9	84.68	85.6	0.92%
10	89.79	86.1	3.69%
11	92.18	94.6	2.42%
12	96.41	96.1	0.31%
13	92.83	95.4	2.57%
14	95.05	96.1	1.05%
15	80.56	76.1	4.46%
16	82.79	75.9	6.89%
17	67.91	71.8	3.89%
18	88.23	91.6	3.37%
19	93.09	90.8	2.29%
20	94.54	91.2	3.34%

Therefore, the proposed algorithm meets the required accuracy of IRI measurement and offers a highly efficient, automatic, and feasible method to measure pavement roughness. In addition, it is more convenient to conduct and cheaper than traditional measurement method.

7. Conclusions

An integrated measuring system based on wireless sensors was developed to enhance the efficiency of pavement roughness measurement, which comprised a data collector, car mounted terminal, and processing system.

Real time data are collected by multiple sensors including accelerometers, GPS, and angular accelerometers. The circuit and case for the data collector were designed to minimize its size. A ZigBee wireless transfer module was adopted for in-vehicle data transmission, and data collectors were located above both left and right rear wheels in car to improve measurement accuracy. A car mounted terminal was designed to package and transmit the data, integrating acceleration data, location information, and vehicle driving information. A shared server was established to receive the collected data, along with in-house software, also compiled in C++, to analyze the uploaded data, calculate IRI with A PSD analysis method, display the results, and generate reports.

Typical sections of road arteries in Huzhou were selected for field testing. Outcomes from the proposed system were compared with actual RQI, verifying the accuracy (maximum relative error below 10%) and the efficiency and stability of the proposed system.