Achieving Environmental Performance Goals - Evaluation of Impact Factors Using a Knowledge Discovery in Databases Approach

(1)

Peer-review under responsibility of the scientific committee of the 23rd CIRP Conference on Life Cycle Engineering doi: 10.1016/j.procir.2016.03.108

Procedia CIRP 48 ( 2016 ) 230 – 235

ScienceDirect

23rd CIRP Conference on Life Cycle Engineering

Achieving Environmental Performance Goals - Evaluation of Impact

Factors using a Knowledge Discovery in Databases Approach

Patrick

Dehning

a

*, Klara Lubinetzki

a

, Sebastian Thiede

b

, Christoph Herrmann

b

a

Volkswagen AG, Postbox 011/1897, Wolfsburg 38436, Germany b

Chair of Sustainable Manufacturing and Life Cycle Engineering, Institute of Machine Tools and Production Technology (IWF), Technische Universität Braunschweig, Langer Kamp 19b, 38106 Braunschweig, Germany

* Corresponding author. Tel.: +49-5361-9-85537; fax: +49-5361-957-18265. E-mail address: patrick.dehning@volkswagen.de

Abstract

In recent past stakeholders have increasingly turned their attention to the environmental performance of companies. This is due to the impact manufacturing has on the environment e.g. climate change or the contamination of soil, air and water. Therefore many companies aim to act responsible and set themselves targets for environmental improvements. Thus they have to measure the performance in terms of energy and fresh water savings or the reduction of waste, volatile organic compounds released and greenhouse gas emissions. This paper aims to support decision makers and corporate management to analyze impact factors influencing a company’s environmental performance and therefore to evaluate the possible risks for not achieving the targets set. A knowledge discovery in databases (KDD) approach is applied for an analysis within the automotive industry to determine these influences.

Peer-review under responsibility of the scientific committee of the scientific committee of the 23rd CIRP Conference on Life Cycle Engineering.

Keywords: Environmental Performance, KDD, Impact Factors, Key Performance Indicators, Automotive Manufacturing

1.Introduction

The usage of diminishing natural resources and the contamination of soil water and air are of rising concern to society, governments and industrial leaders. The rising population and harmonization of living standards are main reasons for an increasing global footprint [1]. Consequentially, consumption of goods and services is rising. Alongside, an elevated request for personal transportation over the past years challenges the automotive industry in satisfying customer demand. It can be seen by the rising number of vehicle sales worldwide from about 39 million cars in 1999 to 71 million in 2014 [2]. To address the environmental challenges in a competitive market companies set themselves targets for reducing their environmental impact and accordingly production cost. Examples for companies having set themselves such goals are the BMW Group and the Volkswagen group, who aim

to reduce e.g. disposal waste and fresh water consumption per vehicle, compared to a defined base line year [3, 4]. A multitude of improvement measures for the production sites are considered and implemented to achieve the targets set. Also impact factors, like plant size, weather conditions or utilization can influence the attainment of goals in a positive or negative way. Identifying these impact factors and quantifying their influence can support management to take appropriate steps to achieve the environmental targets set. This paper presents an approach to evaluate possible impact factors influencing environmental key performance indicators of the vehicle production by analyzing different automotive production plants worldwide and thereby supports management to recognize possible risks for target achievement.

(2)

2.State of research

In this section the state of research is addressed in the matter of environmental performance measurement and the corresponding influencing factors.

2.1. Environmental Performance Measurement

Many scholars worldwide discuss approaches to measure and improve the environmental performance of entities. System boundaries are varying from in house comparison of a single company to measuring the performance of countries and the whole world. One example for a worldwide comparison is the “Environmental Performance Index” that ranks countries by their improvement in different areas considering 20 indicators [5]. According to the ISO 14001 environmental performance on corporate level can be measured against the organization’s environmental objectives or policy using indicators [6]. The ISO 14031 defines an environmental performance evaluation as a process of selecting indicators, collecting and analyzing data using environmental performance indicators (EPIs). The ISO 14031 sets guidelines for selecting EPIs but does not define a mandatory set due to varying structures, sizes and product portfolios of organizations [7].

Many authors address the topic of environmental performance indicators and define and categorize EPIs for various application scenarios. Wagner describes EPIs as quantitative and qualitative means to measure environmental performance of companies using a case of pulp paper and electricity industry considering 14 different EPIs [8]. Jasch defines EPIs as a method to monitor and trace environmental performance enabling benchmarking and reporting on company level [9]. Scholars regard EPIs as a profound way to measure environmental performance within specific boundaries, but the impact factors influencing a company’s EPIs are often neglected or just considered to limit scale[8, 9, 10].

2.2. Evaluation of Influencing Factors

The way environmental performance of a company is addressed in literature varies among authors. Some highlight the importance of an environmental management system (EMS), due to mandatory targets set and the need to plan the implementation of an EMS [10]. Others are focusing on the positive impacts of pull factors like customer and stakeholder demand on the implementation of cleaner production technologies. Blok et al. analyzed the role of regulations, policy and user behavior for promoting a sustainable future [11]. Govindan et al analyzed the factors driving the development of green manufacturing within companies by using a fuzzy approach evaluating mainly external non metric influences [12].

A more detailed view on influencing parameters on plant level was presented by Boyd who developed an energy performance indicator (EnPI) [13]. The EnPI is calculated by using a stochastic frontier approach and takes into account multiple parameters influencing the energy intensity of manufacturing plants. For the automotive

industry product size, weather conditions and utilization as main influences were considered, not quantifying the extent of influence [13].

The literature presented has shown that most authors are concerned with external barriers and drivers promoting a shift to environmental conscious manufacturing. Impact factors directly influencing a company’s EPI are not addressed in depth, with the exception of energy intensity. Therefore the following section introduces an approach to evaluate the factors directly influencing a multitude of environmental performance indicators of manufacturing companies.

3.Knowledge Discovery in Databases (KDD) Approach

Environmental management and reporting, including the use of EPIs, is of increasing relevance in manufacturing companies [8, 14, 15]. In this context environmental data is collected on different levels within a company to validate target achievement, to control environmental performance and to report the current status [3, 4]. A multitude of available data presents an additional opportunity that changes in a company’s EPIs can be reasoned on by impact factors. A KDD approach can be employed to identify the relevant impact factors using a multivariate analysis. It is defined as “[…] the nontrivial process of identifying valid, novel, potentially useful, and ultimately understandable patterns in data” [16]. The KDD process used in this research is based on the process presented by Fayyad et al.. It is divided into six different steps and was extended using an impact analysis [16]. These steps can be seen in Figure 1.

Figure 1: Process steps of the KDD process

The first and most important step is the data selection, because it is the basis for all of the following steps. Data used in the process needs to be reliable, valid and relevant [17]. After selection the target data requires to be processed. Strategies for processing include treating missing values by assigning new values, guessing a value, reselecting the data or to omit wrong values or the whole variable data set due to insufficient data [17].

A qualitative examination is recommended to further reduce Knowledge

1 •Data selection

2 •Processing

3 •Impact analysis

4 •Preliminary data analysis

5 •Data Mining

6 •Interpretation

• Selecting data

• Checking for validity, reliability and relevance

• Treatment of missing values

• Treatment of outliers

• Identification of key impact variables

• Checking for relevant statistical information

• Using statistical methods like regression, classification and clustering

(3)

the number of datasets and choose appropriate datasets according to the analysis’ target [18]. The impact analysis first presented by von Reibnitz can be used to show interactions among variables through expert knowledge and improve the interpretation of further analysis’ results [19]. The following preliminary data analysis is suggested to be done before a multivariate analysis takes place to identify a possible need for transformation and to identify possible clusters [17]. Data mining involves the application of various statistical methods. Among these are classification, regression and clustering methodologies [16]. In this paper the Partial Least Square Regression Analysis (PLS) is employed to identify knowledge on environmental performance indicators [20]. It was chosen because of its high exploratory focus and possible application for various dependent variables [21, 22]. Completing the analysis process with the interpretation of detected patterns and structures can be considered the generation of new knowledge. This analysis step ensures that the data derived is transformed into useful information and is checked for validity and reliability [16].

4.Case Study

This case study will employ the presented KDD approach to generate knowledge on the impact factors on EPIs for automotive manufacturing plants. Data from manufacturing plants worldwide is used provided by the Volkswagen group and analyzed using the software R.

4.1.Evaluation of Influencing Factors

An internal database is used to centrally gather and store the group’s environmental data. Within this database 118 manufacturing plants are available which enter up to 70 different indicators defined by an internal standard on quarterly or annual basis. Due to the amount of data and the scope of this paper only the key performance indicators (KPIs) are considered within the analysis. The considered indicators can be seen in Table 1.

Table 1: Key Performance Indicators considered

Indicator Description Unit

Energy Consumption of electrical, heat energy and fuel gases per unit

kWh/unit

CO2 Total emitted carbon dioxide

(total CO2) per unit

kg/unit Waste Hazardous and non-hazardous

waste meant for disposal per unit

kg/unit

Water Total fresh water consumption per unit

m³/unit

VOC Volatile organic compound (VOC) emissions per unit

kg/unit

Following the selection of indicator data, the impact factors are selected. Impact factors are variables that may influence the further change of KPIs taken into consideration. In total 13 impact factors are identified and analyzed employing the impact analysis. Further impact factors are neglected due to insufficient data quality, e.g. prices (water, energy, waste),

degree of automation and average age of used production equipment. Therefore some possible impact factors are not considered in this approach.

4.2.Impact analysis

The impact analysis is applied to have a first estimation of possible results through expert knowledge and to compare the KDD results to them. Assigning strength of possible influence through the impact analysis is dominated by subjective perception by awarding scores from zero to two for each impact factor; zero meaning no influence, one meaning the same influence and two higher influences. The analysis was conducted jointly with five company experts in form of a workshop to ensure different perspectives are accounted for. The chosen impact factors can be categorized in four categories production planning factors (PPF), product factors (PF) and equipment factors (EF) which are internal factors and external factors (EXF) that are often determined by a plants location. The impact factors along with their impact analysis score can be seen in Table 2

.

Table 2: Impact factors considered

Category Label Designation Unit Score

PPF VUNITS/

CUNITS

Production units vehicles/ components 14 PF GROUND Average to be painted surface m² 13 PPF CAPUTIL Capacity utilization % 12 PPF EEMPLOYE Amount of environmental employees at plant No. of employees 8 PPF EMPLOYE Amount of employees at plant No. of employees 8

EF SIZE Plant size m² 7

EF AGE Age of plant Years 7

EXF HGT Heating degree

days Absolute value 5 EF VCAP/ CCAP Production capacity vehicles/ components 3

EXF WATERS Water stress

index

[1-10] 3

EXF EMISSION CO2-Emission

factor for electrical energy

kgCO2/MWh 3

4.3.Preliminary data analysis

Within the preliminary data analysis the applied data is checked for outliers and normal distribution to find possible groups of plants and restrict the balance area. This is done by using the QQ-plot as an instrument employing the software R. An example of a QQ-plot is given in Figure 2.

(4)

Figure 2: CO2 QQ-plot

One can see that data points show different slopes when trespassing a certain level. The graphical analysis leads to the conclusion that grouping of data points can lead to more accurate results due to the vicinity to normal distribution [22]. Cross validation showed that these groups are basically the plants type (component, vehicle or mixed). All the QQ-plots of the individual groups have shown a vicinity to normal distribution. Thus these three groups will be analyzed separately in the next process steps. Also outliers can be identified easily through the graphical analysis and deleted from the model to improve results.

4.4. Data mining and interpretation

The Partial Least Square Analysis is applied on the identified groups to analyze the factors presented for their direction of influence. It is important to keep in mind, that not all factors can be used for the identified groups, e.g. CCAP as an influencing factor cannot be employed for vehicle plants. The direction of influence is described by the negative or positive sign. On the one hand a positive sign means a rising KPI and therefore is not desired by the company. On the other hand a negative sign signifies a reduction of the KPI and is therefore a desired influence. Also the quality of the regression is checked by employing

the coefficient of determination R2 which is an indicator

how well the regression fits the actual curve [17]. As an example for the three groups the results of the coefficient analysis are shown for the mixed plants only. In the knowledge discovery a summary of all findings will be

presented. For CO2 the PLS regression coefficients for

mixed plants are shown in Figure 3. The regression itself has an acceptable R² of 0.7, whereas the individual coefficients have high standard errors. It can be seen that only two indicators show clear direction of influence. These are the painted vehicle ground (GROUND) which leads to

an increase in CO2 emission per vehicle and the vehicles

produced (VUNITS) that is decreasing the emissions per vehicle. The painted ground as a size measure can be reasoned through the higher amount of energy needed within the processes, e.g. in paint ovens.

Figure 3: Regression coefficients for the CO2 KPI for mixed plants

For the energy KPI of mixed plants the coefficients calculated are shown in Figure 4. The regression used here has an R² of 0.83 and has therefore a high fit. Here a clear trend can be seen which impact factors are influencing a negative development for the indicator and promote thereby a reduction in energy usage per product. These influences are the vehicle capacity (VCAP) and the produced vehicles (VUNITS). In these cases it can be reasoned that a base load of energy is needed to produce the first product and the base load is divided by each additional vehicle produced. The factor promoting an increase of the energy KPI is the number of heating degree days (HGT) for mixed plants. That can be explained by the heating required to keep working conditions in the hall on an acceptable level.

Figure 4: Regression coefficients for the energy KPI for mixed plants

The coefficient results for the waste KPI for disposal considering only mixed plants can be seen in Figure 5. The R² in this case is 0.95 and the regression has therefore a high fit. The results suggest mostly positive impacts especially for vehicles produced (VUNITS), utilization Theoretical Quantiles Sam ple Quantiles 0e +00 2e+05 4e+05 6e+05 8e+05 1e+06 -2 -1 0 1 2

Normal distribution line

Possible groups SIZE EMPLOYEE EEMPLOYEE VUNITS CUNITS VCAP CCAP E EM CAPUTIL EEE AGE GROUND HGT HGT HGT WATERS EMISSION Coeff icien ts Coefficient influence Standard error Variables SIZE EEMPLOYEE VUNITS CUNITS VCAP CCAP CAPUTIL AGE GROUND HGT H TGTT WATERS EMISSION Coeff icien ts Coefficient influence influence Standard error L L EMPLOYEE Variables

(5)

(CAPUTIL) and emission factor (EMISSIONS). On the one hand the influence of VUNITS and CAPUTIL can be reasoned by more attached departments and their corresponding waste output. On the other hand the emission factor has no obvious influence on the waste KPI that could suggest a statistical error.

Figure 5: Regression coefficients for the waste KPI for mixed plants

For the VOC KPI both mixed plants and vehicle plants showed similar behaviour and thus the results shown in Figure 6 are transferable to vehicle plants as well.

Figure 6: Regression coefficients for the VOC KPI for mixed plants

The results suggest that the painted ground (GROUND) has a positive impact on the VOC KPI that is due to the vehicle size and the amount of paint needed. Water stress (WATERS) on the other hand has a negative impact on the VOC KPI, a possible explanation for that is the need to switch to dry scrubbing in paint shops which saves water and lessens VOC emissions at the same time.

For the mixed plants no conclusive results could be derived for the water KPI because of the high standard error of the individual coefficient. Therefore the direction of influence

for the considered factors could not be determined using the PLS method.

All analyses have shown considerable high standard error for most of the impact factors considered. Therefore the direction of influence could only be determined for certain factors. Reasons for that can be the diversity of the balance areas with different degrees of vertical integration, no impact on the analysed KPI or impact factors not yet considered within the PLS approach. Possible solutions to decrease the standard error are the restriction of balance areas and the elimination of factors with unclear direction of impact.

4.5. Knowledge discovery and risk evaluation

The PLS has shown influencing factor and the direction of their influences which can result in possible risks for target achievement of automotive companies. A summary can be seen in Figure 7 for the separate groups individually for each indicator with their direction of influence. Factors with high standard errors resulting in an undefined direction of influence are not mentioned.

Figure 7: Summary of PLS results

The impact factors with the highest impact are discussed further, these are namely UNITS/CAPUTIL, GROUND, HGT, WATERS, EMISSION and AGE.

UNITS/CAPUTIL promotes an improvement of the energy

KPI and therefore of the CO2 KPI due to the per unit

calculation of the indicators and the mandatory base load of energy. This was expected by the impact analysis as well. GROUND indicates that bigger vehicles are connected to higher KPIs, especially in mixed production sites. This impact was expected through the impact analysis. Decisions on product developments should therefore go in line with distance to target achievement considerations to avoid unexpected KPI increases.

HGT bears threats for mixed production sites when facing cold climate. The impact analysis did not identify this factor as a main influence which was due to the assumption that it

has an impact only on the KPIs energy and CO2.

WATERS has as many positive as negative impacts, meaning that a risk can arise from this factors volatility,

UNI TS GROUND CAPUTI L EEM PLOYEE EM P L OYEE SI ZE AGE CAP HGT WA TERS EM ISSI O N Component -4 -4 -4 4 Vehicle 4 -4 8 Mixed 4 -4 4 Component -4 4 -4 4 Vehicle -4 4 -4 4 -4 Mixed -4 4 -4 4 4 Component -4 -4 4 Vehicle -4 -8 8 Mixed Component -8 4 -4 -4 Vehicle 8 -4 4 8 Mixed -4 -4 -4 Component Vehicle 8 4 Mixed 4 -8 14 13 12 8 8 7 7 5 3 3 3

Impact analysis score CO2 Energy Water Waste VOC SIZE EMPLOYEE E EEMPLOYEE VUNITS CUNITS VCAP CCAP CAPUTIL AGE GROUND HGT WATERS EMISSION Co eff icien ts Coefficient influence influence Standard error Variables SIZE VUNITS CUNITS VCAP CCAP SIZE SIZE CAP CAP CAPUTIL AGE GROUND HGT WATERS EMISSION Coeff icien ts Coefficient influence Standard error EEMPLOYEE AG EMPLOYEE Variables

No influence Suggested decrease of the KPI Suggested increase of the KPI

(6)

because it can be influenced by the company’s location choice. Also the water stress index’ influence on the CO2

KPI cannot be reasoned and could therefore be a statistical error.

EMISSION has the strongest and almost always negative impact, which means that either the emission factors need to be decreased or that further influences should be considered which are increasing along with the emission factor. This effect was not foreseen through the impact analysis and the possibility of statistical errors should not be neglected as well.

AGE symbolizes that especially older plants benefit from acquired expertise and retrofits for the waste and water KPIs. Also it should be considered that newer plants are often build in developing countries where waste management systems are not as well established as in the industrialized countries and is therefore not treated in proper way. AGE bears a threat for the KPI VOC, because older plants are still applying older technologies in the paint shop with higher VOC emissions.

The risk arising from all those factors is that changes in the impact factor behavior may lead to stagnations or downfalls in the KPI improvements or even failures to reach overall goals. Therefore location selection, quantity and type of products produced, the necessary amount of employees, the size of a plant and increasing resource scarcities have to be considered in managerial decision making.

5.Summary and outlook

This paper has presented a methodology to evaluate possible impact factors on the environmental performance of manufacturing companies through a KDD approach. Employing a multivariate analysis on the example of an automotive company has shown several impact factors on the chosen environmental KPIs with their direction of influence. Thus the PLS method was able to demonstrate that both external and internal impact factors have to be considered for achieving environmental goals. While directions of influence of the available impact factors were presented within this paper, the exact strength of their influence remains unclear due to the level of uncertainty of coefficients. For future analysis the uncertainty and standard error could be coped with by restricting the balance area furthermore and by improving the amount and quality of available data.

References

[1] United Nations Publication Fund. State of World Population 2011. People and possibilities in a world of 7 billion. New York: United Nations Publication Fund, 2011. State of world population. 2011. [2] Scotiabank. Number of cars sold worldwide from 1990 to 2015 (in

million units). [online] [viewed 28 September 2015, 12:00]. Available from:

http://www.statista.com/statistics/200002/international-car-sales-since-1990/.

[3] BMW Group. Sustainable Value Report 2014, 2015. [4] Volkswagen AG. Volkswagen Sustainability Report 2013.

Wolfsburg, 2014.

[5] Hsu, Angel, Jay Emerson, Laura Johnson, Omar Malik, Jason D. Schwartz, Marc A. Levy, Alex de Sherbinin, and Malanding Jaiteh. The 2014 Environmental Performance Index. New Haven, 2014. [6] ISO. ISO. ISO 14001:2015, ISO 50001:2011 Energy management

systems - Requirements with guidance for use.

[7] ISO. ISO. ISO 14031:2013, ISO 14031:2013 Environmental management -- Environmental performance evaluation -- Guidelines.

[8] Wagner, Marcus. Environmental Performance and the Quality of Corporate Environmental Reports: The Role of Environmental Management Accounting. In: Pall M. Rikhardsson, Martin Bennett, Jan Jaap Bouma, and Stefan Schaltegger, eds. Implementing Environmental Management Accounting: Status and Challenges: Springer Netherland, 2005.

[9] Jasch, Christine. Environmental and material flow cost accounting. Principles and procedures. [New York], [Vienna, Austria]: Springer, 2009. Eco-efficiency in industry and science. v. 25.

[10] Iraldo, Fabio, Testa, Francesco, and Frey, Marco. Is an

environmental management system able to influence environmental and competitive performance? The case of the eco-management and audit scheme (EMAS) in the European union [online]. Journal of Cleaner Production. 2009, 17(16), 1444-1452. Available from: 10.1016/j.jclepro.2009.05.013.

[11] Blok, Vincent, Long, Thomas B., Gaziulusoy, A. Idil, Ciliz, Nilgun, Lozano, Rodrigo, Huisingh, Donald, Csutora, Maria, and Boks, Casper. From best practices to bridges for a more sustainable future: advances and challenges in the transition to global sustainable production and consumption [online]. Journal of Cleaner Production. 2015. Available from: 10.1016/j.jclepro.2015.04.119. [12] Govindan, Kannan, Diabat, Ali, and Madan Shankar, K. Analyzing

the drivers of green manufacturing with fuzzy approach [online]. Journal of Cleaner Production. 2015, 96, 182-193. Available from: 10.1016/j.jclepro.2014.02.054.

[13] Boyd, Gale. Development of a performance-based industrial energy efficiency indicator for automobile assembly plants, 2005. [14] Ammenberg, J., and Hjelm, O. The Connection Between

Environmental Management Systems and Continual Environmental Performance Improvements [online]. Corporate Environmental Strategy. 2002, 9(2), 183-192. Available from: 10.1016/S1066-7938(02)00011-8.

[15] Comoglio, Claudio, and Botta, Serena. The use of indicators and the role of environmental management systems for environmental performances improvement: a survey on ISO 14001 certified companies in the automotive sector [online]. Journal of Cleaner Production. 2012, 20(1), 92-102. Available from:

10.1016/j.jclepro.2011.08.022.

[16] Fayyad, Usama, Piatetsky-Shapiro, Gregory, and Padhraic, Smyth. From Data Mining to Knowledge Discovery in Databases. AI Magazine; Vol 17, No 3: Fall 1996; 37. 1996, (3).

[17] Chatfield, Christopher and Alexander J. Collins. Introduction to multivariate analysis. London, New York: Chapman and Hall, 1980. [18] Beavers, Amy S., Lounsbury, John W., Richards, Jennifer K., Huck,

Schuyler W., Skolits, Gary J., and Esquivel, Shelley L. Practical considerations for using exploratory factor analysis in educational research. Practical assessment, research & evaluation. 2013, 18(6), 1-13.

[19] Reibnitz, Ute von. Szenario-Technik. Instrumente für die unternehmerische und persönliche Erfolgsplanung. 2. Aufl. Wiesbaden: Gabler, 1992.

[20] Wehrens, Ron. Chemometrics with R. Multivariate data analysis in the natural sciences and life sciences. Heidelberg, New York: Springer, 2011. Use R!

[21] Wold, Svante, Sjöström, Michael, and Eriksson, Lennart. PLS-regression: a basic tool of chemometrics [online]. Chemometrics and Intelligent Laboratory Systems. 2001, 58(2), 109-130. Available from: 10.1016/S0169-7439(01)00155-1.

[22] Chin, Wynne W. How to Write Up and Report PLS Analyses. In: Vincenzo Esposito Vinzi, Wynne W. Chin, Jörg Henseler, and Huiwen Wang, eds. Handbook of Partial Least Squares. Berlin, Heidelberg: Springer Berlin Heidelberg, 2010, pp. 655-690.