PARALLEL MODEL BASED FAULT DETECTION ALGORITHM FOR ELECTRONIC PARKING BRAKE SYSTEM

1. INTRODUCTION

The most severe test condition for the parking brake is that it must keep the vehicle stationary on the 30 percent road slope in gross vehicle weight states. Around 40 kgf of human power is required for the hand-operated lever type brake, and 50 kgf for the foot-operated pedal type brake (Moon et al., 2002). However, sometimes drivers, such as females or seniors, cannot supply sufficient operational power. Thus, cases have been reported in which accidents occurred after parking. To prevent these, the electronic parking brake (EPB) system, which generates parking force through a simple switch operation by the driver, is required.

Currently produced EPB can be largely classified into two types. The first is a cable puller type that uses motor to pull parking cable as shown in Figure 1, where the latter is a motor on caliper type that generates braking force by delivering the power directly from motor to caliper.

The cable puller type is divided once again into either a single puller type or a dual puller type depending on the structure of system. In the structure of single puller type,each cable is connected to the actuator and then two cables are connected through the equalizer. In contrast, two cables are directly connected to the actuator in the structure of dual puller type. Note that the single puller type is easier to install on a vehicle compared to the dual puller type. The single puller type is used in the EPB system of this paper, where its structure is shown in Figure 2 (Chung et al., 2008).

Figure 1. Configuration of the cable puller EPB system

Figure 2. Single puller type EPB system

For an EPB system to supply sufficient parking force, a parking force sensor should measure the force of the EPBsystem. If a fault occurs in this sensor, sufficient parking force may not be supplied, thereby seriously threatening the safety of the vehicle. Thus, a fault detection method is required for the parking force sensors of EPB systems to improve the safety of vehicles. For this purpose, a highly reliable model based fault detection method is needed to detect abnormal fault signals, which cannot be detected by the existing on-line sensor monitoring fault detection methods (Han et al., 2008). In the last decade of the 20th century, the field of fault detection has shown rapid progress due to safety demands in automotive industry. Besides, fault detection has become a must for many other industries due to productivity and quality considerations (lsquo;zero defectsrsquo; manufacturing) (Albas et al., 2001).

The existing online sensor monitoring fault detection methods are limited because they only check some measurable output variables. Since the conventional approaches do not provide deeper insight and usually do not allow fault detection, several model-based fault detections are developed by using input/output signals and applying dynamic process models. The model-based fault detection algorithm predicts the future values of the states, outputs, and comparing them with the measured values (Sreedhar et al., 1995).

The model-based fault detection method can easily determine the defection in the system as shown in Figure 3. Residual is the difference between the actual system and the designed model. If there is no fault in the system and the model is precisely designed as the real system, the error value would be zero. However, the difference is not be zero in practical applications because the model is not based on the complete physical data (Isermann and Balle, 1997;

Figure 3. Block diagram of the model-based fault detection algorithm.

Figure 4. Structure of the multi model-based fault detection algorithm

Figure 5. Structure of the parallel model-based fault detection algorithm.

Pfeufer, 1997). In the previous studies (Boskovic and Mehra, 1998; Maybeck and Stenvens, 1991; Gopinathan et al., 1998), the multi model-based fault detection algorithm has been widely used as shown in Figure 4. In order to detect faults existing in one sensor among a quantity of n sensors, the proposed algorithm uses the rest of n-1 number of sensors to generate n-1 number of fault detection models. Note that the minimum number of sensors to apply this algorithm is three in this case. The reason for implementing many models for one sensor is that the possibility of a false alarm is high and it needs more time to determine faults if fault detection is conducted by using one model.

In this paper, the parallel model-based fault detection algorithm consisting of the mathematical model, the fuzzy model, and the neural network model is proposed to improve the reliability of fault detection as shown in Figure 5. The proposed algorithm can be applied to a system, where only one sensor could be implemented into the model. In addition, the adaptive thresholds and the operation counting method are applied to enhance the robustness against the system modeling errors, disturbances, and noises. This paper is organized as follows. In Section II, modeling methods of three independent models are presented and the parallel model-based fault detection algorithm for EPB system is proposed. The proposed algorithm is verified by hardware-in-the-loop (HIL) simulation in Section III and concluding remarks follow in Section IV.

2. PARALLEL MODEL-BASED FAULT DETECTION ALGORITHM FOR ELECTRONIC PARKING BRAKE

2.1. Basic Structure To apply the parallel model-based fault detection algorithm for EPB system, three independent models – a mathematical model, a fuzzy model, and a neural network model – should be developed to estimate parking force using four input variables – an operating time, a voltage, a current, and a temperature. In the mathematical model, a parity space method (Patton and Willcox, 1987; Beard, 1971) is used

剩余内容已隐藏，支付完成后下载完整资料

PARALLEL MODEL BASED FAULT DETECTION ALGORITHM FOR ELECTRONIC PARKING BRAKE SYSTEM

1. INTRODUCTION

The most severe test condition for the parking brake is that it must keep the vehicle stationary on the 30 percent road slope in gross vehicle weight states. Around 40 kgf of human power is required for the hand-operated lever type brake, and 50 kgf for the foot-operated pedal type brake (Moon et al., 2002). However, sometimes drivers, such as females or seniors, cannot supply sufficient operational power. Thus, cases have been reported in which accidents occurred after parking. To prevent these, the electronic parking brake (EPB) system, which generates parking force through a simple switch operation by the driver, is required.

Currently produced EPB can be largely classified into two types. The first is a cable puller type that uses motor to pull parking cable as shown in Figure 1, where the latter is a motor on caliper type that generates braking force by delivering the power directly from motor to caliper.

The cable puller type is divided once again into either a single puller type or a dual puller type depending on the structure of system. In the structure of single puller type,each cable is connected to the actuator and then two cables are connected through the equalizer. In contrast, two cables are directly connected to the actuator in the structure of dual puller type. Note that the single puller type is easier to install on a vehicle compared to the dual puller type. The single puller type is used in the EPB system of this paper, where its structure is shown in Figure 2 (Chung et al., 2008).

Figure 1. Configuration of the cable puller EPB system

Figure 2. Single puller type EPB system

For an EPB system to supply sufficient parking force, a parking force sensor should measure the force of the EPBsystem. If a fault occurs in this sensor, sufficient parking force may not be supplied, thereby seriously threatening the safety of the vehicle. Thus, a fault detection method is required for the parking force sensors of EPB systems to improve the safety of vehicles. For this purpose, a highly reliable model based fault detection method is needed to detect abnormal fault signals, which cannot be detected by the existing on-line sensor monitoring fault detection methods (Han et al., 2008). In the last decade of the 20th century, the field of fault detection has shown rapid progress due to safety demands in automotive industry. Besides, fault detection has become a must for many other industries due to productivity and quality considerations (lsquo;zero defectsrsquo; manufacturing) (Albas et al., 2001).

The existing online sensor monitoring fault detection methods are limited because they only check some measurable output variables. Since the conventional approaches do not provide deeper insight and usually do not allow fault detection, several model-based fault detections are developed by using input/output signals and applying dynamic process models. The model-based fault detection algorithm predicts the future values of the states, outputs, and comparing them with the measured values (Sreedhar et al., 1995).

The model-based fault detection method can easily determine the defection in the system as shown in Figure 3. Residual is the difference between the actual system and the designed model. If there is no fault in the system and the model is precisely designed as the real system, the error value would be zero. However, the difference is not be zero in practical applications because the model is not based on the complete physical data (Isermann and Balle, 1997;

Figure 3. Block diagram of the model-based fault detection algorithm.

Figure 4. Structure of the multi model-based fault detection algorithm

Figure 5. Structure of the parallel model-based fault detection algorithm.

Pfeufer, 1997). In the previous studies (Boskovic and Mehra, 1998; Maybeck and Stenvens, 1991; Gopinathan et al., 1998), the multi model-based fault detection algorithm has been widely used as shown in Figure 4. In order to detect faults existing in one sensor among a quantity of n sensors, the proposed algorithm uses the rest of n-1 number of sensors to generate n-1 number of fault detection models. Note that the minimum number of sensors to apply this algorithm is three in this case. The reason for implementing many models for one sensor is that the possibility of a false alarm is high and it needs more time to determine faults if fault detection is conducted by using one model.

In this paper, the parallel model-based fault detection algorithm consisting of the mathematical model, the fuzzy model, and the neural network model is proposed to improve the reliability of fault detection as shown in Figure 5. The proposed algorithm can be applied to a system, where only one sensor could be implemented into the model. In addition, the adaptive thresholds and the operation counting method are applied to enhance the robustness against the system modeling errors, disturbances, and noises. This paper is organized as follows. In Section II, modeling methods of three independent models are presented and the parallel model-based fault detection algorithm for EPB system is proposed. The proposed algorithm is verified by hardware-in-the-loop (HIL) simulation in Section III and concluding remarks follow in Section IV.

2. PARALLEL MODEL-BASED FAULT DETECTION ALGORITHM FOR ELECTRONIC PARKING BRAKE

2.1. Basic Structure To apply the parallel model-based fault detection algorithm for EPB system, three independent models – a mathematical model, a fuzzy model, and a neural network model – should be developed to estimate parking force using four input variables – an operating time, a voltage, a current, and a temperature. In the mathematical model, a parity space method (Patton and Willcox, 1987; Beard, 1971) is used

剩余内容已隐藏，支付完成后下载完整资料

资料编号：[137696]，资料为PDF文档或Word文档，PDF文档可免费转换为Word