PARAMETER IDENTIFICATION OF A FIRE TRUCK SUSPENSION FOR VIBRATION ANALYSIS

: In case of field measurement it is essential to determine the measurement conditions. With accurate parameter identification credible measurement and simulation results can be achieved. In this paper the parameter identification of a CSD-755-10 heavy-duty fire truck suspension for vibration analysis is presented, which is an important base-point for further research.


Introduction
In special purpose vehicles for example fire trucks high value equipment is built in. These devices can be damaged by various adverse effects, like harmful vibrations during operation of a fire truck, which must be continuously examined [1]. These vibrations can cause serious structural damage; therefore their examination is an important, current issue [2]- [3]. Analyzing the available literature on the topic it can be stated that similar measurements and tests were mostly carried out in laboratory conditions or in test track [4]- [7], field experiments were performed in a smaller number [8]- [9]. Most experiments were carried out with several different purpose trucks [4]- [9], but not with special fire trucks. Literature connected to firefighting is about firefighter training [10]- [11] or the development of rescue robots [12]. Field experiments can be carried out in a very wide spectrum. The output of the experiments, data analysis and

Parameter identification of vehicles
Parameter identification is not a simple task since the appropriate method should be determined by knowing the task to be performed and ensuring the necessary conditions. Each test, measurement can be extremely complex and varied, therefore parameter identification has several well-known methods. In practical applications statistical parameter identification methods are widely used. In [13]- [14] a maximum likelihood estimation method is presented to identify the parameters of a bus based on Auto-Regressive-Moving-Average (ARMA) and Auto-Regressive-Moving-Average model with eXogenous inputs (ARMAX) models. In [15] a mathematical model to identify the stiffness and damping of a heavy-duty truck suspension system is presented based on curve fitting method. Using curve fitting method the minimum of quadratic error function between the simulated seat vertical acceleration power spectral density and the measured power spectral density was taken. Publication [16] concerns exploitational identification of structural parameters of the Toyota Camry suspension system based on measured acceleration data. The developed algorithm combines the restoring force, boundary perturbation and direct parameter estimation techniques. In [17] a modified recursive least squares algorithm was adjusted with the forgetting factor and Kalman filter was used to identify the parameters of a motorcycle suspension system. Paper [18] presents a novel online method for identifying the mass of an off-road vehicle using acceleration measurements of sprung and unsprung mass using recursive least squares estimation. The developed method can be used for vehicle active safety control, traction control, and powertrain control. In [19] a novel parameter identification method for vehicle tracking using an extended Kalman filter and measurements of road surface vibrations is presented. In [20] a fast continuous time system identification is described using the half-vehicle model and a prediction error based method direct continuous time system identification to evaluate the vehicle's unknown parameters. The developed method was tested with simulation software CarMaker. In [21] a novel algebraic estimator is presented to find the unsprung mass of the suspension system. The purpose of this research was to improve passenger comfort with a sliding mode controller in simulation environment. In [22] the parameters of a heavy-duty truck engine mount are identified based on measurement data through optimization with Least Mean Square (LMS) algorithm. The main objective of the research was to develop an adaptronic engine mounting system in order to enhance driver comfort and handling of a heavy truck. In [23] the air temperature of the truck cabin was modeled with white-box and black-box techniques. For black-box modeling Matlab system identification toolbox was used, which estimates the parameters of the system with a least squares algorithm. With the same method the resistance of an electric vehicle battery was also identified using Simulink Parameter Estimation Toolbox [24]- [25]. Not only the parameters of a suspension system can be obtained with identification methods, but the characteristics of certain elements too in order to develop vehicle simulations. In [26] the damping characteristic of shock absorber of a heavy-duty truck suspension system was determined with the least squares method. Seven parameters of a Besinger model describing the shock absorbers nonlinear characteristics were identified based on measurements. Similar methods for parameter and model identification can be used in other engineering fields too like magnetic hysteresis modeling [27], [28] and for linear approximation of nonlinear systems [29].
In case of parameter identification the input and output of a system is known, but the behavior of the system is unknown. The aim is to establish a model that can fit to the measured data. If the system is entirely unknown it is called the 'black-box' parameter identification as it can be seen in Fig. 1.

Fig. 1. Parameter identification
The task of the research is to examine the vibrations caused by the road during operation of the CSD 755-10 heavy-duty fire truck with measurements and simulation.
To establish an accurate simulation model the parameters of the system need to be determined. For parameter identification the least squares parameter estimation method was chosen. The advantage of the method is that it can be easily implemented and the time domain measurements can be directly used. According to literature study it is one of the most common methods for parameter identification in vehicle development.

Least squares parameter estimation
In case of parameter identification the measurement data and the input signal are given. The behavior of the system can be described by mathematical equation with system variables. An optimization task is to be solved, which objective function is to minimize the sum of least squares [30], (1) To solve the optimization task Matlab OPTI Toolbox was used. As solver NL2SOL adaptive nonlinear least squares algorithm was chosen, which is one of the best algorithm available [31].

Measurement
In Hungary, especially in case of low class roads, the participants of the traffic encounter a lot of surface defects and potholes. As first step the depth of several potholes around Győr was measured, the maximum depth was 10 cm. At the same time the height of the curbs was also measured, which was also 10 cm. Considering the above, it was decided to set up a 10 cm high artificial obstacle on which the vehicle would drive through.
The construction of CSD-755-10 fire truck suspension system can be seen in Fig. 2. The vehicle has a ladder frame chassis with beam axles. Suspension is provided by leaf springs for both axles and auxiliary leaf springs are installed in the rear axle because of constant load. Hydraulic shock absorbers are installed on both sides provide right road holding. The drive system is powered by a CUMMINS 6CT 8.3 diesel engine that drives the FP-2418 pump through a separate gear drive. The special firefighter equipment, which consists of a water tank, a foam tank and storing compartments is assembled to the chassis of the vehicle with a separate auxiliary chassis. The cabin is placed on rubber mount.  The measurement was carried out on a flat, straight, solid pavement. After installing the accelerometer sensors 10x10 cm obstacles were placed on the selected road section on which the vehicle had to drive through at constant velocity during the test, which was specified as 5 km/h. 5 km/h speed was chosen because it is the smallest speed that can be kept by the vehicle, and the tire and the chassis are not damaged during the jump. This was considered the starting point for the measurements. Two measurements were recorded. Both measurements were considered appropriate. From the acceleration data velocity and displacement data was calculated using integration. The measurement results can be seen in Fig. 4. The measured signals were noisy; therefore prior to parameter identification the noise not related to the measurement should be filtered out. Noise filtering was carried out with Maple. First the frequency of the noise was determined by a fast Fourier transformation then a low pass filter was used to remove excess noise. Using the results remaining after the noise filtration computer simulation was carried out.

Modell
In the literature quarter vehicle model was mostly used for parameter identification. Therefore in the first approach, the parameters of the front axle were determined using a linear and nonlinear quarter vehicle model presented in Fig. 5. The parameter identification of the fire truck using a half or full vehicle model will be the task of further research.
Equations describing the nonlinear characteristics of the spring and the damper are: The obstacle was modeled as a step function. During parameter identification, several models were used. First a linear model without tire damping was applied, since according to the literature the damping of the tire is very small compared to the other parameters and is therefore neglected in some cases [32]- [34]. Next, the effect of tire damping was observed; taking this into account a better approximation could be achieved. Comparing the measured and simulation results a remaining sinusoidal signal was found. It was assumed that the sinusoidal signal could be generated from the road or from the vehicle, so the model was extended with a sinusoidal disturbance signal. The test models used during the identification were as follows: 1. Linear model without tire damping; 2. Linear model; 3. Nonlinear model; 4. Linear model with sinusoidal road disturbance; 5. Linear model with sinusoidal vehicle disturbance; 6. Nonlinear model with sinusoidal vehicle disturbance.

Parameter identification with initial parameter values taken from literature
For the first approximation, the initial parameter values were taken from [35] as it is shown in Table I. The results of parameter identification are shown in Table II. The numbers show the test models described before and ΣLS is the sum of least squares.
From Table II it can be concluded that the best approximation could be achieved by using a nonlinear model with sinusoidal noise from the vehicle (model 6). However, it can be stated that the spring stiffness was very low compared to the initial values and the other models. The damping of the tire was greater than the value obtained from the literature in each case.

Determining spring stiffness with other methods
The stiffness of the leaf spring was determined with other methods too. Based on an analytical solution described in [36] the spring stiffness was calculated as k=522845.5 N/m. This method has an advantage that only the geometry of the spring is necessary for calculation.
The spring stiffness was obtained by finite element simulation too [37]. The finite element model can be seen in Fig. 6. A remote displacement was constrained at the eyes and a force was given at the top leaf [38]. This is a simplified model as inter-leaf friction was not taken into account; contacts were modeled as no separation.
Parameter identification was carried out with pre-determined spring stiffness values. In this case less unknown parameters had to be obtained. Results can be seen in Table III. Comparing the results it can be stated that the best fit was achieved with the nonlinear model, which takes the vehicle noise into account (model 6). At the same time it should be noted that, unlike expectations with less parameters better fit could not be achieved.

Summary of results
To summarize the results it can be concluded that the best fit could be achieved with the nonlinear model with sinusoidal vehicle noise. Compared to the parameter values taken from the literature the damping effect is greater. The stiffness of the tire was less than the value taken from the literature in all cases. From the nonlinear models it can be assumed that the leaf spring has progressive and the tire stiffness and the hydraulic shock absorber has degressive characteristics.
Comparing the methods it can be stated that in case of the analytical solution and the finite element analysis the geometric data is sufficient, it is not necessary to carry out expensive and time-consuming measurements, which is a great advantage. However, it should be noted that these are approximate models in which some external effects are neglected. Analyzing the data of the measurements, it can be seen that an unexpected sinusoidal effect emerged. This is an important observation as on the theoretical plane its consequences could not be detected.

Conclusion and further research tasks
It can be concluded that accurate models and field measurements are indispensable for parameter identification and creating simulations for practical usage. With the experience gained during the present research, even more accurate measurements will be carried out in the future for more effective parameter identification.
In addition to the presented results, the recent research work has helped to identify future research tasks, which are parameter identification in the frequency domain, validation of the model based on measured data and using half-vehicle or full vehicle models for simulations.