DEVELOPMENT OF REGRESSION MODEL FOR DRIVING FATIGUE DETECTION BASED ON SEAT PRESSURE DISTRIBUTION FORCE OF THE DRIVERS

(1)

DEVELOPMENT OF REGRESSION MODEL FOR DRIVING FATIGUE DETECTION BASED ON SEAT PRESSURE DISTRIBUTION FORCE

OF THE DRIVERS

Ani M.Firdaus

Universiti Teknikal Malaysia Melaka Kamat S.Rahayu

Associate Professor of Manufacturing Engineering

Faculty of Manufacturing Engineering, Universiti Teknikal Malaysia Melaka seri@utem.edu.my

Minhat M.

Doctor of Manufacturing Engineering

Faculty of Manufacturing Engineering, Universiti Teknikal Malaysia Melaka mohdm@utem.edu.my

Husin K.

Associate Professor

Faculty of Education, International Islamic University College kalthom@kuis.edu.my

ABSTRACT

The aim of present work is to develop of the regression model for a biomechanical factor that is seat pressure distribution force of the drivers. Ten subjects participated in this study. The pressure distribution force responses were taken and evaluated using the Tekscan CONFORMAT pressure map.

The seat pressure distribution force is one of the contributors to the drivers’ fatigue problem, which led to road accidents among Malaysian. The process modelling using regression analysis was proposed to formulate and develop the regression model which can estimate the relationship between the process input parameters and output response. Design Expert 8.0.6 software was used for the regression analysis. The regression model was successfully developed and validated. The results of the validation show that the validation runs were within the 90% prediction interval of the developed model. Besides, the residual errors were less than 10% as compared to the predicted values. The significant parameters have been identified in this study; time exposure, type of road, gender, the interaction between time exposure and gender, and interaction between type of road and gender, which influenced the pressure distribution force. The contribution of this study is obvious as the resulting outcomes can be capitalized as guidelines to the development of regression model. In future, the author suggests that the more study on developing regression model should be focused as the published work on the application of ergonomics is too lacking.

Keywords: Biomechanical, driving fatigue, regression model, seat pressure distribution force.

Introduction

Rather than the risky driving, and speeding, one of the leading cause of the road accidents is fatigue.

Driving fatigue is a subject that is getting increasing attention in the road safety field. Fatigue in terms of driving is a feeling of tiredness and reduction of alertness which is associated with drowsiness, and which impairs capability and willingness to perform the driving task [San T P. P. et. al, 2016]. The drivers should care about fatigue as in the short period fatigue cause discomfort, and reduced motor control and strength capacity [Björklund M. et.al, 2000]-[Huysmans M. A. et. al, 2010]. In term of long period, fatigue’s effect can lead to performance decrement, lower productivity, lower quality of work, and raised the risk of incidents and human error [Marcus Yung].

(2)

A comfortable seat while driving plays a significant role in influencing the performance and quality of driving. Many previous studies had proved that the seat pressure distribution force gives a significant effect to driving fatigue. Satoru et al, studied the relationship between the driver’s posture and behavior change as driving for a long period and the changes in body pressure distribution between the driver and seat [Furugori S. 2003]. They clarified that the change in body pressure distribution and the change of drivers is a good index to predict fatigue. The main finding is when the time changes, the pressure had devoted on one point of the center part of the back. The pressure begins to increase in the area of the devoted part. The results from the analysis show that there is a potentiality to consider driving fatigue caused by the pressure distribution.

Another study compared the effect of pressure distribution force on the car seat when driving through different road conditions [Rahayu S., Firdaus M., 2003]. This study calculates the contact area of the seat pressure distribution force. They found that the winding road condition has the higher contact area for both genders as the maximum pressure distribution force is higher compared to other road conditions. They claimed that the area of the contact pressure influenced the pressure distribution produced on the seat. As the pressure distribution becomes higher, the driver tends to be fatigue while driving.

In a recent study by [Tayeb I., 2016], they proposed the method to recognize fatigue and drowsiness state by analysis of pressure distribution on the seat. They measured changes in the position of the drivers during high activity and over long periods of time via analyzing the changes in pressure force over time. All these previous studies showed that the seat pressure distribution gives a significant effect on fatigue. Indirectly, caused the driver to involve in road accidents that increases every year.

Based on statistic data, 7,152 road deaths and 521, 466 accidents have been reported in 2016 [Department of Statistic Malaysia, 2016][Malaysia Institute of Road Safety Research, 2017][Ministry of Transport Malaysia, 2017]. Besides, the previous study has predicted that there will be 10,716 deaths caused by accidents in 2020 [Sarani R. et.al, 2012].

Hence, this study counters this problem by investigating the seat pressure distribution force that contributes for driver fatigue among Malaysian while driving at the high-risk accidental area and different road conditions, and finally develop the regression models as the fatigue prediction model. In recent year, the development of fatigue prediction model has received significant attention and focus from the researcher in the fields of aeronautics, transportation, excavating and professional sports.

This current study attempts to develop the regression models of seat pressure distribution force by using the regression analysis in order to reduce the risk of fatigue during driving and minimize the road accident’s number in Malaysia.

Methods

1. Subject and Population

There were 10 (5 males and 5 females) healthy and experiences drivers engaged as the subjects for the study. These subjects represent the 95th percentiles populations. This selection of population based on the previous study by [Hassan S. N. Et. al, 2015] as the Malaysian’s mean statute height and standard deviation for males and females were 169.57±7.57 cm and 156.83±5.97 cm respectively, which the range is in 95th percentiles population. Besides, the subjects were engaged based on their driving experience, which is at least two years of driving experiences.

2. Test Apparatus and Protocol

As this study involves the Malaysian population, and Malaysia road system and environment, the author had chosen Malaysia’s national car, Proton Saga FLX 1.3L (automatic transmission) as the test vehicle for this study. Proton Saga is categorized as the affordable or economic cars in terms of their price and usage. This statement is proven with the awards received by this car such as 2011 Best People’s Car, 2011 Best Value for Money Family Car Award, and 2010 Best Passenger Car Model of

(3)

alcohol, smoking, and free from taking any type of medicines. The subjects were given some explanation about the experiment and the map of the road to be taken. The Tekscan CONFORMat pressure map was used to measure the pressure force that occurs between the driver and the seat interface. The tekscan sensor senses the pressure distribution force between a subject and seat cushion in real time, shows the information as color-coded real-time display and records the data as a movie for users to review and analysis. Figure 1 shows the Tekscan CONFORMat pressure map attached to the seat cushion of the test vehicle.

!

Figure 1: Sensor placed on the seat cushion 3. Response Surface Modelling and Validation

The response surface modelling (RSM), which suitable for modelling and optimization was carried out in this current study for the process of the model’s development and validation. The regression analysis by RSM in form of the polynomial equation is used to define the relationship between the input parameters and the output response. The Design Expert 8.0.6 software was used for this development. The quantitative validations was used to validated the final regression as the validation runs must meet this two following conditions; first the model can predict the validation run outcome within 90% of its predictive interval (PI), second, the accuracy of a process model can be assessed by using residual error method, which the error should be less than 10% to represent the accuracy of the model. [Vaughn N. A., Polnaszek C., 2007].

Result and Discussion

This part presents the findings and discussion for this study, which consist of two section; first section is the development of regression models, and second section is validation of the models for seat pressure distribution force.

1. Regression Modelling of Seat Pressure Distribution

The data from experimental runs data were recorded in Table 1. Three factors were considered as the input factors; type of road, gender, and time exposure. While the mean values of seat pressure distribution force (kPa) as the response (output) in this study.

(4)

Table 1: Input factor’s data and output response in this study

Std Run

Factor 1 Factor 2 Factor 3 Response 1

A: Time Exposure, min B: Type of Road C: Gender Pressure Distribution Force, kPa

2 1 15.00 Straight Female 11.800

26 5 15.00 Winding Female 11.963

19 9 15.00 Uphill Female 12.400

27 10 15.00 Uphill Female 12.233

23 11 30.00 Uphill Female 12.933

32 12 30.00 Uphill Female 12.500

31 13 15.00 Downhill Female 12.133

8 17 15.00 Straight Male 13.650

4 18 15.00 Straight Male 13.550

25 19 30.00 Straight Male 13.875

18 20 30.00 Straight Male 13.820

21 21 15.00 Winding Male 14.175

9 22 15.00 Winding Male 14.250

28 23 30.00 Winding Male 14.300

7 24 30.00 Winding Male 14.540

5 25 15.00 Uphill Male 13.500

15 26 15.00 Uphill Male 13.600

3 27 30.00 Uphill Male 13.633

13 28 30.00 Uphill Male 13.533

6 29 15.00 Downhill Male 14.300

29 30 15.00 Downhill Male 14.200

(5)

The analysis continues with the determination of appropriate polynomial equations to represent the relationship between the input parameter and output responses by conducting the sum of squares sequential model (SMSS) and lack of fit test as reflected in Table 2 and Table 3 respectively.

Based on the results, show that the relationship between factors and output response can be modeled using 2FI (factor interaction) equations. This results based on the F-Value for the 2FI smaller than the others. Besides, the p-value for 2FI is greater than F-Value.

Table 2: SMSS analysis for seat pressure distribution force model

Table 3: Lack of fit test for the seat pressure distribution force model

3. ANOVA for Response Surface 2FI Model

Based on ANOVA analysis as represent in Table 4, the Model F-value of 67.81 implies the model is significant. There is only 0.01% chance that a Model F-value this large could occur due to noise.

Besides, the accuracy of this model by the lack of fit analysis as the “Lack of Fit F-value” of 0.26 indicates that, the lack of fit is not significant relative to the pure error, and there is an 85.21% chance that a “Lack of Fit F-value: this large could occur due to noise.

Source Sum of

Squares df Mean Square

F Value

p-value Prob >

F Mean vs

Total 5490.05 1 5490.05

Linear vs

Mean 27.08 5 5.42 58.20 < 0.0001

2FI vs

Linear 1.75 7 0.25 7.04 0.0003 Suggested

Quadratic

vs 2FI 0.000 0 Aliased

Residual 0.67 19 0.035

Total 5519.55 32 172.49

Squares df Mean Square

F Value

p-value Prob > F

Linear 1.78 10 0.18 4.43 0.0042

2FI 0.031 3 0.010 0.26 0.8521 Suggested

Quadratic 0.031 3 0.010 0.26 0.8521 Aliased

Pure Error 0.64 16 0.040

(6)

The p-value of less than 0.1 indicates that the model terms are significant. In this case, time exposure, type of road, gender, the interaction between time exposure and gender, and interaction between type of road and gender are significant model terms. These significant factors influenced the effect of seat pressure distribution forces of the subjects.

Table 4: ANOVA analysis of the 2FI model

4. Polynomial Equation

Based on surface response modeling the 2FI polynomial equation developed to relate the input parameters to the seat pressure distribution force is present in Table 5 for the equation in terms of actual factors.

Squares df Mean

Square

F- Value

p-value Prob

> F

Model 28.83 12 2.40 67.81 < 0.0001 significant

A-Time

Exposure 0.61 1 0.61 17.14 0.0006

B-Type of

Road 1.78 3 0.59 16.79 < 0.0001

C-Gender 24.69 1 24.69 696.92 < 0.0001

AB 0.021 3 7.020E-00

3 0.20 0.8963

AC 0.13 1 0.13 3.73 0.0685

BC 1.59 3 0.53 14.99 < 0.0001

Residual 0.67 19 0.035

Lack of Fit 0.031 3 0.010 0.26 0.8521 not

significant

Pure Error 0.64 16 0.040

Cor Total 29.50 31

(7)

Table 5: Polynomial equation for the seat pressure distribution force.

5. Validation of Regression Model

These final regression model were been validated to check whether the developed regression model can predict the hand grip pressure. The validation results of three sets of parameter setting are shown in Table 6. These sets of parameters are selected through the point prediction capability by Design Expert software and using an independent data. The results indicate that actual seat pressure distribution force data from the validation runs fall within the 90% prediction interval and the residual errors are ranging from 0.905% to 1.724%, which in absolute value is less than 10%. Hence, this model is accurate enough to predict the resultant seat pressure distribution force within 90% CI and the residual error relative to predicted values are less than 10%.

Table 6: Validation runs for regression model of seat pressure distribution

Conclusion

This study has shown that, the regression model is successfully formulated and established to relate the relationship between the seat pressure distribution force input process parameters; time exposure, type of road, and gender, and an output response; seat pressure distribution force. The study found that the significant parameters and interaction factors are time exposure, type of road, gender, the

Parameter Equation

Type of Road Straight

Pressure Distribution = +11.19875 + 0.026533 * Time Exposure

Gender Female

Type of Road Straight

Pressure Distribution = +13.51225 + 9.40000E-003 * Time Exposure

Gender Male

Type of Road Winding

Gender Female

Type of Road Winding

Gender Male

Type of Road Uphill

Gender Female

Type of Road Uphill

Gender Male

Type of Road Downhill

Gender Female

Type of Road Downhill

Gender Male

Input Parameters

Prediction (kPa)

90% PI low (kPa)

90% PI Hi (kPa)

Actual

(kPa) Error (%) Time

Exposure

Type of

Road Gender

15.00 Straight Female 11.600 11.211 11.983 11.800 1.724

22.50 Straight Female 11.796 11.432 12.160 11.935 1.178

30.00 Downhill Male 14.370 13.980 14.751 14.230 0.905

(8)

interaction between time exposure and gender, and interaction between type of road and gender. These significant parameters and interaction factors, which identified by ANOVA analysis gives the significant influence on the seat pressure distribution force. Besides, the study successfully validated the regression model as the runs data fall inside the 90% PI and the residual error less than 10%. Thus, the regression models can predict the values of seat pressure distribution force, and assists the users to be more alert by detecting the early indication of fatigue. As a results, minimize the road accidents and fatalities in Malaysia. The author believes that these findings make several noteworthy contributions to the development of regression models for detection fatigue especially in transportation field.

Acknowledgements

The author would like to give an appreciation to Universiti Teknikal Malaysia Melaka (UTeM) and Tokushima University, Japan who gave the technical and financial support.

REFERENCES

San T P. P., Ling S. H., Chai R., Tran Y. EEG-based Driver Fatigue Detection using Hybrid Deep Generic Model // IEEE 38th Annual International Conference of the Engineering in Medicine and Biology Society (EMBC). 2016. PP. 800-803.

Björklund M., Crenshaw A. G., Djupsjöbacka M., Johansson H. Position Sense Acuity is Diminished Following Repetitive Low-Intensity Work to Fatigue in a Simulated Occupational Setting // European Journal of Applied Physiology. 2000. Volume 81(2000). No. 5. PP.361-367.

Côté J. N., Raymond D., Mathieu P. A., Feldman A. G., and Levin M. F. Differences in Multi-Joint Kinematic Patterns of Repetitive Hammering in Healthy, Fatigued and Shoulder-Injured Individuals //

Clinical Biomechanics. Volume 20 (2005) No.2.

Huysmans M. A., Hoozemans M. J., van deer Beek A. J., de Looze M. P., van Dieën J. H. Position Sense Acuity of the Upper Extremity and Tracking Performance in Subjects with Non-Specific Neck and Upper Extremity Pain and Healthy Controls // Journal of Rehabilitation Medicine. Volume 42(2010) No.9. PP. 876-883.

Marcus Yung. Fatigue at the Workplace: Measurement and Temporal Development // Available at:

http://hdl.handle.net/10012/10119.

Furugori S., Yoshizawa N., Iname C., Miura Y. Measurement of Driver's Fatigue based on Driver's Postural Change // Sice 2003 annual conference. Volume 1(2003).PP. 264-269.

Rahayu S., Firdaus M.. Measurement of Driver's Fatigue based on Driver's Postural Change // Sice 2003 annual conference. Volume 1(2003).PP. 264-269.

Tayeb I., Jemai O., Zaied M., Amar C. B Towards a Smart Car Seat Design for Drowsiness Detection Based on Pressure Distribution of the Driver’s Body // ICSEA 2016. 2016. PP. 230.

Department of Statistic Malaysia. Social Statistic Bulletin, Malaysia 2016 // (Online), Available at:

<https://www.dosm.gov.my/v1/index.php> (Accessed: December 15, 2016).

Malaysia Institute of Road Safety Research. General Road Accident Statistic in Malaysia 2016 //

Available at: https://www.miros.gov.my/1/page.php?id=364 (Accessed: September 29, 2017).

Ministry of Transport Malaysia. Transport Statistic Malaysia 2016 // Ministry of Transport Malaysia, Available at: http://www.mot.gov.my/en/resources/yearly-statistic (Acessed: September, 29, 2017).

Sarani R., Rahim S. A. S. M., Marjan J. M, Voon W. S. Predicting Malaysian Road Fatalities for Year 2020 // Malaysian Institute of Road Safety Research. Kuala Lumpur 2012.

Hassan S. N., Yusuff R. M., Zein R. M., Hussain M. R., Selvan H. K. T. Anthropometric data of Malaysian workers // New Ergonomics Perspective: Selected papers of the 10th Pan-Pacific Conference on Ergonomics, Tokyo, Japan, 25-28 August 2014, CRC Press. 2015. PP.353.

Proton. Saga FLX // Available at: https://www.proton.com/ (Accessed: November, 14, 2015).

Wickens C. D., Lee J. D., Y. Liu, Gordon Becker S. E. An Introduction to Human Factors Engineering // Instructor 2009.

Vaughn N. A., Polnaszek C. Design-Expert® software // Stat-Ease, Inc, Minneapolis. MN.2007.