A new Online Hall Effect Sensor Fault Detection and Location in Brushless DC Motor Based on Normalized Phases Currents Analysis

In this paper, a new online technique for Hall Effect sensor fault diagnosis in brushless DC (BLDC) motor is proposed. The proposed technique is based on phase current waveform analysis and does not need any Hall sensor information. The normalized phases current values are analyzed per and post-sensor fault in every cycle. Using a definition of suitable conditions and threshold values for normalized currents values, all sensor fault types (i


INTRODUCTION 1
High efficiency, high torque density, low-level noise, and wide speed control range are the main benefits of BLDC motors related to the induction motors.Therefore, BLDC motors can be one of the best options for many applications, such as robots, electric bicycles, and computer equipment [1][2][3].But torque ripple [3,4] and dependency on the rotor position are the main drawbacks of the BLDC motors.The switching pattern of a BLDC motor driver for speed control is highly dependent on the position sensors output values.The most common BLDC motor position sensors are Hall Effect sensors.These sensors may be damaged during operation of BLDC motors which can cause serious damage to the motor.Therefore, using an online sensor fault and driver faults *Corresponding author email: arehpanahi@tafreshu.ac.ir (M.Arehpanahi) diagnosis technique is very important for the safe operation of the BLDC motor.There are many diagnosis techniques for detection of the sensor faults in BLDC motors.The sensor fault is detected based on the sum of the instantaneous of Hall output signal values with an investigated diagnosis table reported in literatu [5].Cheshta et al. [5] focussed on the analysis of the output values of hall sensors for detection of faulty sensor.A new technique based on a binary combination of Hall signals values and commutation times between power switches of the BLDC driver is presented by Qian, and Ming [6].A direct redundancy-based method by utilizing redundant Hall-effect sensors for Fault Tolerant Control (FTC) of a BLDC motor is presented by Aqil and Hur [7].The online sensor fault diagnosis of BLDC motor are carried out by wavelet package [8], Goertzel Algorithm [9] and improved ZOA (Zeroth Order Algorithm) technique [10].The DC-link current second harmonic component monitoring is employed for sensor fault diagnosis [11].Mehta et al. [12] used analysis of the output sensor signals values sequences as binary numbers, the sensor faults can be detected.In healthy conditions, the sequence of three Hall-Effect sensors values (0 for OFF state and 1 for ON state) is expressed as one of the six binary numbers (from 001 to 110).Therefore, any sequences outside this range will be detected as a fault [13,14].No-detection at low-speed operation, is the main drawback of the mentioned techniques.Tashakori and Ektesab [13] analyzed the output sensor waveform based on reference frame theory and Vector-Tracking Observer (VTO) is employed for detection of faults especially at low-speed operation.But complex calculation process and requiring a rotating harmonic vectors models, are the main disadvantages of work reported in literature [15].A new technique based on line-voltage monitoring using FFT analysis was proposed by Donato et al. [16].Detection process complexity and no-detection during start-up are the main disadvantages of it.An improved FTC scheme based on FDP (Fault Detection Probability) and VTO for sensor fault diagnosis was proposed by Donato et al. [17].Nofault detection during start-up is the main drawback of the reported data [15].Diagnosis of Hall-Effect sensors faults based on output sensors signals values combined with the measured line voltages using Discrete Fourier Transform (DFT) was reported by Mousmi et al. [18].Analysis of the stator phase currents using Stockwell Transform (S-Transform) for hall sensors faults diagnosis was defined by Gowtham et al. [19].In general, no-fault detection at low-speed operation or false diagnosis in transient conditions is two challenges of the mentioned techniques.The inability to separate between starting and fault conditions is a main drawback of the most online sensor fault diagnosis techniques.If any sensors fail, they will affect on the stator current waveform directly.The stator current monitoring is a usual and accessible technique for analysis of the motor behavior.Therefore, in this paper, a new online technique based on phase current waveform analysis is proposed.In the proposed technique, using all phases' current waveforms analysis, the sensor fault can be detected and located effectively.The main contribution of the proposed technique is definition of some conditions for normalized three phase currents values with a simple calculation process.The proposed technique, can able to separate transient conditions from fault conditions.This paper is divided into four sections: section 1 proposed method, section 2 simulation results and discussion, section 3 experimental results and section 4 conclusion.

PROPOSED METHOD
Usually, during sensor failure time, the output value of the sensor is unchanged (0 or 1).Therefore, in this paper, a constant value of the sensor output is considered as a fault.Figure 1 shows a block diagram of the BLDC motor driver which is used in this paper.Table 1 shows the switching pattern of the BLDC motor driver under sensor A fault (set to 0).If a sensor output value is to be constant, the status of some switches will be always ON or OFF.Then, in half of the period (positive or negative) the corresponding phase current is close to zero.According to Table 1, in a period, it is clear that   has positive and negative signs but   has negative sign and   has positive sign.Therefore,   and   have a DC component.If this DC component pre and post-fault is well analyzed, the sensor fault could be well detected.DC component of phase currents affects on the average and RMS value of the phase current.
Therefore, in the proposed technique, the average value of the phase current divided by the RMS value of that which is called normalized phase current, is introduced as a fault indicator.The normalized phase current value is expressed in Equation ( 1

I I dt x A B C TI
where ,    are a period, nominal RMS value of phase current and the normalized value, respectively.In healthy conditions, the average value of all phase  (set to 0 or 1), the average value of all phase current will not be zero anymore.In this condition, according to Table 1, the positive or negative half period of phase current (depended on the sensor number failure) is close to zero approximately.Therefore, the absolute value of the corresponding phase current average will be about 50% of phase current RMS value.Therefore, according to Equation ( 1), the absolute normalized value of corresponding phases is close to 0.5.then using comparison of the mentioned normalized value with threshold value (i.e.0.5), the sensor fault can be detected.However, the main problem in fault detection process, is normalized current variation during transient conditions (for example: starting time).To solve this problem, several simulations of different three-phase BLDC motors types (different output powers) during starting time are done.By analysing their results (variation of the all phase normalized currents values), it was found that the absolute normalized phase currents values are no higher than 0.4.Therefore, it can be defined suitable conditions for the normalized phase currents with an appropriate threshold value i.e. 0.4 (which is independent of BLDC motor specification) for fault indicator.The proposed technique, for detection and location of Hall-Effect sensors failure is expressed as six conditions in Tables 2 and 3(for two faults type).The subscript "n" denotes the normalized value.For example, when three phase normalized currents satisfy in   < 0,   < −0.4,  > 0.4 , based on Table 2, the sensor A has failed and set to 0. In diagnosis process, some problems may negatively impact on the fault detection such as rotation direction change, transient   2 or Table 3, it can be used these tables for reverse rotation direction.Consequently, any change in rotation direction does not effect on the proposed technique results at all.The second problem is transient conditions such as starting.In the proposed technique, this problem has been fixed by selecting 0.4 as a threshold value for the fault indicators.The normalized phase currents values in transient condition do not satisfy in conditions that are expressed in Table 2 or Table 3 with this threshold value.

SIMULATION RESULTS AND DISCUSSION
In order to performance analysis of the proposed technique, two BLDC motors types (motors 1 and 2) with different specifications are considered.The specifications of both motors are listed in Tables 4 and 5.
Both motors (motors 1 and 2) have been started under different load conditions (no-load and full-load).Then sensor faults (logic 0 and logic 1) have been applied to all sensors individually when motors reach steady-state conditions.All simulation results have been done in MATLAB/Simulink software.For example, in motor 1, the fault (logic 0) is applied to the sensor A under full   2 and 3).For motor 2, another sensor fault type (logic 1) is applied to sensor B, under no load conditions at time t=0.25 s (Figures 4 and 5).The phase current waveforms and their normalized values waveforms for motor 1 under full load pre and post-fault are illustrated in Figures 2 and 3.According to Figure 3 (  < 0,   < −0.4,  > 0.4) and based on row 1 from Table 2 it is clear that the sensor A has failed (logic 0).The phase currents waveforms for motor 2, under no load pre and post-fault are shown in Figures 4 and 5.According to Figure 5 (  < −0.4,  > 0,   > 0.4) and based on row 2 from Table 3, sensor B has failed (logic 1).Therefore, the proposed technique could detect and locate sensor fault.
The normalized phase currents values of motor1 under no-load, during starting time is illustrated in Figure 6.During starting time 0.015s<t<0.018sbased on Figure  As a result, motor 1 normalized current values during start-up do not satisfy in Tables 2 and 3 conditions.Consequently, the proposed technique is robust against transient conditions, especially during start-up.At lowspeed operation, the phase current waveforms are changed related to the normal speed and contain low order harmonics.These waveforms are similar to the faulty cases.The sensor fault (sensor C, logic 1) is applied to motor 1 at 300 rpm (7.5% of rated speed).The sensor fault (sensor B, logic 1) is applied to motor 2 at 250 rpm (5% of rated speed).The phase current waveforms and their normalized values of motor 1 pre and post-fault are shown in Figures 7 and 8 3. Therefore, sensor C has failed (logic 1).Also, the normalized phase currents values of motor2 (Figure 10) in post-fault time are   < −0.4,  > 0,   < 0.4 which satisfies in row 2 in Table 3.Consequently, sensor B has failed (logic 1).Therefore, the proposed technique could online detect and locate all of the Hall-Effect sensor faults types in transient conditions, no-load, fullload and low-speed operation

EXPERIMENTAL RESULTS
In order to experimentally verify the proposed technique, a prototype BLDC motor is considered (Figure 11).The prototype BLDC motor is designed for analysis of the motor under sensor fault conditions.In other words, the motor driver software is designed to be able to apply sensor faults.Specification of the prototype BLDC motor is listed in Table 6.The BLDC motor starts under noload conditions.In steady-state conditions, sensor A is disconnected from the driver at t=0.4s (logic 0).Threephase currents pre and post fault is illustrated in Figures 12,13 2, sensor A is failed (logic 0).
To analysis of the proposed technique under lowspeed operations, the prototype BLDC motor operates at low-speed i.e. 250 rpm.The sensor fault is applied to the sensor A (logic 0) at t=0.43s.The three phase currents of the BLDC motor are illustrated in Figures 18,19 2, the sensor A is failed (logic 0).Consequently, the proposed technique could detect and locate online all of the sensor faults types at different operation conditions.

This paper presents a new online detection technique for
Hall-Effect sensor faults in brushless DC motors.In the proposed technique, the sensor fault is detected and located online by the normalized phase currents analysis.The proposed technique is also able to detect sensor faults at low speeds.Using a comparison of all normalized phases current values with suitable threshold value, the sensor fault (type and sensor number) is detected and located effectively.The proposed technique, can separate transient conditions such as starting from faulty cases which is the main advantage of the proposed technique related to the other diagnosis techniques.The simulation results for two different types of the BLDC motors under different loads and speed conditions, show the capability of the proposed technique in sensor fault diagnosis.Verification of the proposed technique is carried out by experimental results effectively.

Figure 1 .
Figure 1.Three-phase BLDC motor driver and low-speed operation.The impacts of these problems on the proposed technique are discussed in this paper.The rotation direction of three-phase BLDC motor is reversed by a shifting any two phases from three phases of the terminal voltages.Therefore, by exchanging only two rows in Table

Figure 5 .Figure 6 .
Figure 5. Normalized phase currents of moto 2 under noload and sensor B fault (logic 1) . The phase current waveforms and their normalized values of motor 2 pre and post-fault are shown in Figures 9 and 10 .

Figure 8 .Figure 9 .Figure 10 .
Figure 8. Normalized phase currents (motor 1) at low speed (300 rpm) and 14.According to the phase currents waveforms and the normalized phase currents which are shown in Figures 15, 16 and 17, the normalized phases currents satisfy in

Figure 12 .Figure 13 .
Figure 12.Phase A current waveform pre and post fault

TABLE 1 .
Switching pattern of BLDC motor driver under sensor A fault (logic 0)

TABLE 6 .
BLDC motor specification for testing

Figure 11 .
Set-up of the prototype BLDC motor