Daily Worker Evaluation Model for SME-scale Food Production System Using Kansei Engineering and Artificial Neural Network

(1)

Agriculture and Agricultural Science Procedia 3 ( 2015 ) 84 – 88

Peer-review under responsibility of Jurusan Teknologi Industri Pertanian, Fakultas Teknologi Pertanian, Universitas Gadjah Mada doi: 10.1016/j.aaspro.2015.01.018

ScienceDirect

The 2014 International Conference on industry (ICoA): Competitive and sustainable

Agro-industry for Human Welfare

Daily Worker Evaluation Model for SME-Scale Food Production

System using Kansei Engineering and Artificial Neural Network

Mirwan Ushada

a,*

, Tsuyoshi Okayama

b

, Atris Suyantohadi

a

, Nafis Khuriyati

a

, Haruhiko

Murase

c

a_{Universitas Gadjah Mada, Dept. of Agro-Industrial Technology Jl. Flora, Bulaksumur, Yogyakarta 55281, Indonesia} b_{Ibaraki University, College of Agriculture, 3-21-1, Chuuo, Ami, Inashiki, Ibaraki, ZIP 300-0393, Japan}

c_{Osaka Prefecture University, Graduate School of Engineering, 1-1 Gakuen-cho, Naka-ku, Sakai, Osaka, ZIP 599-8531, Japan}

Abstract

This paper highlighted a daily worker evaluation model for small medium-scale food production system. The model consist of worker capacity assessment and worker performance evaluation sub-models. The model measures the relationship between Total Mood Disturbance (TMD), heart rate of worker and workplace parameters using Kansei Engineering approach. However, the rapid measurement of TMD is difficult and full of bias since using the paper-based questionnaire of Profile of Mood States (POMS). Therefore, a rapid measurement method was developed using Artificial Neural Network to support the application of daily evaluation model. The inputs of the model were heart rate, workplace temperature, relative humidity, light intensity and noise level, which were measured before and after working. The output was TMD score.The training and inspection data for ANN was collected from workers of food production system as Tempe, Bakpia, Fish Chips and Crackers industries in Yogyakarta Special Region. ANN model were tested successfully predicted TMD score using back-propagation supervised learning method. The trained ANN model generated satisfied root mean square error value. ANN model is possible to substitute conventional data acquisition of POMS. The daily evaluation model is applicable to assist industrial management for providing the appropriate worker assignment for shift schedulling and environmental set point for the workplace comfortability.

Peer-review under responsibility of Jurusan Teknologi Industri Pertanian, Fakultas Teknologi Pertanian, Universitas Gadjah Mada.

Keywords: artificial neural network; daily evaluation; profile of mood states; rapid measurement; total mood disturbance

*_{* Corresponding author. Tel.:+62-274-551-219; fax: +62-274-551-219}

E-mail address: [email protected]

Peer-review under responsibility of Jurusan Teknologi Industri Pertanian, Fakultas Teknologi Pertanian, Universitas Gadjah Mada

(2)

Nomenclature

AH AngerˉHostility LI Light intensity (Lux)

C Confusion NL Noise level (Decibel)

D DepressionˉDejection RH Workplace relative humidity (%)

F Fatigue T Workplace temperature (oC)

FR Friendliness TA TensionˉAnxiety

HR Heart rate (Pulse per minute) V Vigor

1. Introduction

Small medium-scale production system is a fundamental production system in Indonesian industry. Small-medium scale indicated that the amount of worker is from 5 to less than 100 persons (Indonesian Central Agency in Statistics, 2011). There is more than 54.6 million small medium-scale industry who use the man power almost 97.2% (Indonesian Central Agency in Statistics, 2011). There is 26.4 million of small medium-scale food industry in Indonesia. This kind of industry contributes to the Indonesian food sovereignty (Indonesian Ministry of Cooperatives and Small-Medium Enterprises, 2011).

In a production system, worker evaluation is increasingly important since humans are sensible (Nagamachi, 2002; Ushada et al., 2014c). Worker sensibility or known as Kansei is a worker workload response to a different amount of production target which are measurable using the verbal and non-verbal parameters (Nagamachi, 2002; Ushada et

al., 2014b). Evaluating the worker in a food production system is important to decrease error of work method,

achieve high quality product and minimizing production lead time with a good sensibility.

Fig. 1. Daily worker evaluation model for small medium-scale food production system.

This research proposed a daily worker evaluation model for small medium-scale food production system as shown in Fig. 1. The model consist of worker capacity assessment (Ushada et al., 2014a) and worker performance evaluation sub-models (Ushada et al., 2014 b). The model measures a relationship between worker and workplace parameters using Kansei Engineering approach. Kansei Engineering approach is applicable to analyze the worker sensibility using comparison between verbal and non-verbal parameters (Nagamachi, 2002; Ushada et al., 2013).

Total Mood Disturbance Heart Rate Workplace Temperature Relative Humidity Light Intensity Noise Level Worker Performance Evaluation Workplace Maintenance Environment Set Point Before Working After Working Rapid Measurement Daily Check-in/ Check-out Workplace Measurement Worker

Measurement AssignmentWorker Worker Shift Scheduling

Data Logger1 Data Logger2 Data Logger3 Data Logger4 Data Logger5

(3)

Kansei Engineering is a method to extract the verbal and non-verbal response, to help human identifying the needs, preference or achieving satisfaction for product/system (Nagamachi, 1995).

In daily check-in before working and check-out after working, the Total Mood Disturbance (TMD) and heart rate were measured using the paper-based questionnaire of Profile of Mood States (POMS) (McNair et al., 1992). In the same time, workplace temperature, relative humidity, light intensity and noise level were measured. Heart rate were rapidly measured using wrist blood pressure data logger. Workplace temperature and relative humidity were rapidly measured using Thermo Recorder (Extech RH520 Data Logger). Light intensity and noise level were rapidly measured using Luxmeter and Multifunctional Environment data logger. However, the rapid measurement of TMD is difficult since using the paper-based questionnaire of POMS. The paper-based questionnaire was already applied as a calibration method in food production system (Ushada et al., 2013).

Artificial Neural Network (ANN) is applicable for rapid measurement of TMD. It is due to its capability to model large number of non-linear parameters in biosystems and Kansei Engineering (Reed and Marks II, 2002; Ushada et

al., 2014c). Up to date, Ujita and Murase (2006) has developed POMS Software for application of baum test

characteristic. This software was limited for a relationship between non-linguistic emotional display and data of questionnaire method (POMS). This software is not applicable for worker in small-medium scale food production system.

The research objectives are: 1) To propose a daily worker evaluation model for small medium-scale food production system; 2) To develop a rapid measurement model for total mood disturbance using Kansei Engineering and Artificial Neural Network. The research advantage is to support the development of a daily worker evaluation model for small-medium scale food production system. The daily evaluation model is expected to assist industrial management for providing the appropriate worker assignment for shift schedulling and environmental set point for the workplace comfortability as shown in Fig. 1.

2. Materials and Methods 2.1. Case studies

4 (four) cases study of workers in food production systems were evaluated as Tempe, Bakpia, Fish Chips and Crackers. These industries were selected due to the popularity of food product in Indonesian consumer and the unique worker profile.

2.2. Profile of Mood States

Profile of Mood States (POMS) was selected as verbal parameter because it measures the affective mood state fluctuation in a wide variety of populations (McNair et al., 1992; Motohashi et al., 2013; O’Connor, 2014). POMS are proposed to measure five identifiable negative moods or affective states of worker as Tension-Anxiety, Depression, Anger-Hostility, Fatigue, Confussion and two positive moods as Vigor and Friendliness . POMS are measured using questionnaire of 65 words declaring the worker mood. Questionnaire result is analyzed using Total Mood Disturbance (TMD) score in equation 1 (McNair et al., 1992) :

TMD = TA + D + AH + F + C – V- FR (1)

2.3. Artificial Neural Network

A rapid measurement model is defined as a black box relationship between non-verbal parameter of heart rate, workplace environmental parameters and verbal parameter of mood. Its function for predicting TMD score can be defined as dynamic variation of heart rate in a given workplace environmental parameters. Thus, these relationships were modelled using a three layered ANN as shown in Fig.2a. The ANN source code was built using Microsoft Visual Basic Application for Microsoft Excel (Okayama, 2014). The inputs of the model were total heart rate, workplace temperature, relative humidity, light intensity and noise level, which were measured before and after working. The output was TMD score.

(4)

Fig. 2. Artificial neural network model for rapid measurement of TMD: (a) ANN Structure; (b) Learning performance.

3. Results and Discussions

The case study of food production system was Tempe, Bakpia, Fish Chips and Crackers industries in Special Region of Yogyakarta. The data was recapitulated in to 572 data set. The 544 data set were used in training ANN model, while the remains were used in inspection data. Table 1 indicates an input and output range of ANN training data. ANN model was trained using these training data.

Table 1. Sensitivity analysis for ANN model

Range Input Output

HR (pulse per minute) T (o_C) _{RH (%)} _{LI (Lux)} _{NL (Decibel)} _{TMD Score}

Minimum 63 25 30 1 16 4

Maximum 123 46 92 12237 3644 12

Based on the sensitivity analysis of output error by trial and error basis (Table 2), seven neurons in the hidden layer were determined. The architecture of network consisted of five neurons in the input layer, seven neurons in the hidden layer and one neuron in the output layer (5-7-1).

Table 2. Sensitivity analysis for ANN model

No Hidden Neuron Learning

Coefficient Momentum Iteration RMSE

1 5 0.1 0.9 10,000 0.282 2 5 0.1 0.9 20,000 0.278 3 6 0.1 0.9 10,000 0.277 4 6 0.1 0.9 20,000 0.277 5 7 0.1 0.9 10,000 0.274 6 7 0.1 0.9 20,000 0.277 7 7 0.2 0.8 20,000 0.2691 8 8 0.1 0.9 10,000 0.274 9 8 0.1 0.9 15,000 0.277 1_{Minimum error}

The training converged after approximately 20,000 iterations, learning coefficient of 0.1 and momentum of 0.9 (Fig. 2b). The root mean square error values of training and inspection data were 0.269 and 0.208 respectively. ANN model were tested successfully predicted TMD score using back-propagation supervised learning method.

0 0,2 0,4 0,6 0,8 1 0 5000 10000 15000 20000 Ro ot M ea n Square Erro r Learning Iteration RH LI NL HR T TMD

a

b

(5)

4. Conclusions

This paper proposed a daily worker evaluation model for small medium-scale food production system using Kansei Engineering and Artificial Neural Network. The model consists of worker capacity assessment (Ushada et

al., 2014a) and worker performance evaluation sub-models (Ushada et al., 2014b). Worker heart rate, workplace

temperature, relative humidity, light intensity and noise level could be measured daily, before and after working using the available data logger while the total mood disturbance could be measured daily using rapid measurement of profile of mood states. Kansei Engineering is applicable as a new methodology for developing an ANN model as a rapid measurement of total mood disturbance using back-propagation supervised learning method. The trained ANN model generated satisfied root mean square error value. The ANN model is applicable to substitute the conventional data acqusition of paper-based POMS questionnaire. For further improvement it is suggested that the data of POMS are measured precisely minimizing the bias of workplace condition and work methods. The precise data will assist the easier training process and stucture selection of ANN model.

Acknowledgements

We would like to express our sincere gratitute and acknowledge the financial support from Directorate General for Higher Education, Indonesian Ministry of Education and Culture for 2014 International Collaboration Competitive Research Grant for International Publication of Universitas Gadjah Mada No: LPPM-UGM/1008/LIT/2014. We would like to express our sincere gratitute to industries of Tempe, Bakpia, Fish Chips and Crackers. Finally, we would like to acknowledge the support for data acquisition from Arum Baashitu, Reza Purnama Triendy and Dwita Rahmawati.

References

Indonesian Central Agency in Statistics.2011. Data of Small-Medium Scale Enterprises

Indonesian Ministry of Cooperatives and Small-Medium Enterprises. 2010. Strategic Plan of Indonesian Cooperatives and Small-Medium Enterprises (In Indonesian) 2010-2014.

Motohashi. N., Yamaguchi, M., Fujii, T. And Kitahara, Y. 2013. Mood and Cognitive Function following Repeated Transcranial Direct Current Stimulation in Healthy Volunteers: A Preliminary Report. Neuroscience Research. 77(2013) 64-69

McNair, D. M., Lorr, M., Droppleman, L. F. 1992. Manual for the Profile of Mood States (Rev. ed.). San Diego, CA: Educational and Industrial Testing Service.

Nagamachi, M. 1995. Kansei Engineering: A New Ergonomic Consumer-oriented Technology for Product Development. International Journal of Industrial Ergonomics. Vol 15(1995), 3-11.

Nagamachi, M. 2002. Kansei Engineering as A Powerful Consumer-oriented Technology Development. Applied Ergonomics. Vol 33(2002), 289-294.

O’Connor, P. J. 2004. Evaluation of Four Highly Cited Energy and Fatigue Mood Measures. Journal of Psychosomatic Research. 57(2004) 435-441

Okayama, T. 2014. Neural Network Macro. Microsoft Visual Basic Application for Microsoft Excel. Microsoft Inc Japan, 2014.

Reed, R. D. and Marks II, R. J. 1999. Neural Smithing, Supervised Learning in Feed Forward Artificial Neural Networks. Massachusetts Institute of Technology Press, 1999, 49-95.

Ujita, Y, Murase, H. 2006. Profile Of Mood States Software for application of baum test characteristic. Proceedings of the 3rd International Symposium on Machinery and Mechatronics For Agriculture and Biosystems Engineering (ISMAB) 23-25 November 2006, Seoul, Korea Ushada M., Mulyati, G. T., Guritno, A. D., Murase, H. 2013. Combining Drum-Buffer-Rope algorithm and Kansei Engineering to control

capacity constrained worker in a bioproduction system in 2013 International Federation of Automatic Control (IFAC) Biorobotics Conference, Japan Paper No. 11, 2013.

Ushada, M, Okayama, T, Murase, H. 2014a. Development of Kansei Engineering-based Watchdog Model To Assess Worker Capacity In Indonesian Small-Medium Food Industry. Engineering in Agriculture, Environment and Food. Status: Under Review

Ushada, M, Okayama, T., Suyantohadi, A., Khuriyati, N., Murase, H. 2014b. Kansei Engineering-based Artificial Neural Network Model to Evaluate Worker Performance in Small-Medium Scale Food Production System. International Journal of Industrial and Systems Engineering. Status: Submitted

Ushada, M, Okayama, T., Suyantohadi, A., Khuriyati, N, Murase, H. 2014c. Combining Kansei Engineering and Artificial Neural Network to Assess Worker Capacity in Small-Medium Food Industry in Proceeding of International Conference on Electrical Engineering, Computer Science and Informatics (EECSI 2014), Yogyakarta, Indonesia, 20-21 August 2014