Abstract
Introduction
The exceptionally high cost of occupational and industrial accidents for organizations and nations globally is alarming. Over 264 million industrial accidents with over 350,000 fatalities occur each year, which means that each day approximately 1,000 people die as a result of occupational accidents [1]. Most epidemiological data concerning musculoskeletal disorders relate to developed and industrialized countries, but there is little information on this subject in developing and low-income countries [2], causing losses greater than 20 billion annually [3]. The high complexity of manufacturing systems in terms of interactions between humans and their working environment continuously provides challenging problems for researchers and communities alike [4].
As more effort is invested in the development of standards and guidelines for the design of workplaces and the production of machines and equipment to be used and serviced by humans, data from the working population has become increasingly relevant [5]. Several of the directives and standards developed by CEN, e.g., those dealing with machine safety and personal protective equipment depend on anthropometric information, which is lacking for Serbia despite the very high number of occupational accidents in the metal industry [6]. In order to meet the most urgent need for such data from Serbia we undertook the present study.
An ergonomics approach to the design of an industrial workstation attempts to achieve an appropriate balance between the worker capabilities and work requirements so as to optimize worker productivity and the total system, as well as to provide worker physical and mental well-being, job satisfaction and safety. The physical dimensions in the design of an industrial workstation are of major importance from the viewpoint of both production efficiency and operator well-being [7]. An improperly designed workstation causes static muscle efforts, resulting in acute localized muscular fatigue, and consequently in reduced performance and productivity, with enhanced risk of operator related health hazards [8]. Small changes in workstation dimensions can have a considerable impact on worker productivity and occupational health and safety. The authors in [9] confirm that muscular fatigue is common in workplaces in China’s industries. The highest prevalence of muscular fatigue is for low back (54%), neck (43%), and shoulders (42%) [10] and an ergonomically redesigned workstation is an effective intervention program to reduce fab workers’ awkward shoulder postures and shoulder symptoms [11]. In addition, manufacturing machines should be customized to fit the anthropometric specifications of the end users in order to prevent unnecessary ergonomic problems in the future [11].
The importance of studying this problem greatly exceeds the number of published papers in this area. That is the reason why this paper collects data on Serbian metal industry workers and defines critical industrial workers‘ models through multivariate modeling in order to facilitate their accommodation at their workstations. This study is the first ever comprehensive anthropometric study of Serbian metal industry workers in the country, whose labor force is increasingly employed on both the local and international market and by both domestic and foreign multinationals who have set up their subsidiary plants in Serbia. Our survey also verifies the models with a new sample collected from the population of metal industry workers. The multivariate results are also compared to those obtained using percentiles and the proposed methodology is thus proved.
Previous research
Studies of new industrial workers’ anthropometric data in different countries have appeared in recent years, since they are usually lacking or are out of date. The authors in study [12] determined the anthropometric dimensions for Portuguese industrial workers, while those of Filipino workers are examined in study [13]. In order to meet the most urgent need for such data from Norway, the authors in [14] gathered data for workers engaged in light industry and office work. According to survey [15], it is evident that body dimensions differ between students and workers; namely, male students have larger dimensions than male workers. Accordingly, it is very important not to use data pertaining to the general population in design [7].
In addition to the lack of data, there also exists a lack of awareness of the importance of ergonomics in industrial enterprises. Today, the engineers and designers in industry have poor knowledge of both the formal design processes in use in their companies and how to apply ergonomics principles [16], so that the installed designs revealed several serious ergonomics problems which could impact on the operators’ ability to work efficiently and safely.
Workplace layout design parameters or dimensions are determined using the existing anthropometric data, usually by means of the percentiles method [17], or simulations and/or computerized models due to the high costs of real modeling [4]. The traditional percentile approach is criticized by certain authors for the reduction in accommodation when two or more dimensions are involved in a design such as in the case of industrial workplaces [18]. When it concerns percentiles, a common requirement is to fit the 5th percentile to the 95th percentile person, which means the design that fits 90% of the population of interest, using the univariate approach. Also, the 5th and 95th percentile values of a dataset of global anthropometric data are asymmetrically distributed [19]. Accordingly, the authors in [20–23] point out the multivariate approach as the alternative method to the percentile approach, and propose PCA. The authors in [24] even go so far as to claim that the application of univariate approaches to multivariate design problems provide inaccurate estimates ofaccommodation.
The practical problem lies in the fact that percentiles are not additive, and when the design problem requires several dimensions for proper fitting, this problem results in less than 90% of the population fit. For instance, if 90% of the population have a 5th-95th percentile stature, it is possible that only 82% of those 90% have a 5th-95th percentile chest circumference, and only 78% of them a 5th-95th percentile shoulder breadth, and so on [20]. Accordingly, it is not surprising that it was shown that the conventionally determined workspace envelope could not be reached by 95% of the population but rather by only 73% of men and 75% of women [21]. These findings indicate that an alternative method to the percentile approach could be multivariate.
We intend to verify that empirically and will use both percentile and PCA multivariate modeling on the anthropometric data of 172 metal industry workers collected in the Serbian metal industry and to compare them to see the differences.
An anthropometric study of Serbian metal industry workers
Samples
Industry represents one of the most important economic sectors due to its multiple impact on the development of every other production and non-production activity. Throughout the year 2011, out of the total of Serbia’s 7.3 milion population 375,000 workers were employed in Serbia’s industry, of which 104,972 were employed in the metal industry [25]. The proportion of female metal industry workers in the Serbian labor market is very low [25]. The metal industry used to be the backbone of the Serbian economy in the past dececades, but today it is characterized by a substantial number of weak enterprises undergoing the reorganization process in the aim of sustainable and dynamic development in order to fit in to the unique market of the European Union. The expansion of the EU by the joining of new member countries sets a new task before joint industrial policy: the harmonization and compliance of different industrial systems. Therefore the employment growth of a large number of metal industry workers in Serbia is expected by 2020 [26]. Consequently, the emphasis in this study is placed on workplaces in the metal industry, because it is the most promising sector with a growing rate of exports. The upgrading of workplaces in terms of ergonomics opens up avenues for higher worker productivity and other improvedbusiness performances.
Our first sample of metal industry workers comprised 122 and the second control sample included 50 participants. In both samples all the participants are male. The average age of the first sample of participants was 45, with standard deviation of 9.96 years. Different metal industry companies were selected so as to represent the main branches belonging to the Serbian metal industry: the transport industry, information technologies, electronics, agricultural mechanization, the defense industry and shipbuilding. The measurements for the first sample were taken in 33 industrial companies willing to participate in the study located throughout Serbia. The companies were from both the public and private sectors. The research included SMEs (97%) and large-sized (3%) firms. Accordingly, these sample characteristics ensured the intended representativeness.
The sample was formed by means of the static anthropometry method, which implies measuring in the erect position while standing and sitting (so that the torso is at a 90° angle with the upper leg, and the upper leg at a 90° angle with the lower leg). As can be seen in Table 1, a total of 8 basic static anthropometric dimensions and body weight were recorded for each individual, namely stature (mm), seat height (mm), upper leg length (mm), lower leg length (mm), shoulder width (mm), hip width (mm), arm length (mm) and shoe length (mm). In previous works[7, 28] it was shown that these 8 anthropometric measurements are quite sufficient for dimensioning, since they act on the principles of the mechanical mechanism. The standard anthropometric instruments used in this study were an anthropometer, beam caliper, sliding calipers, and steel tape. Other instruments included a weight scale and a stool for seated measurement. The participants remained in their working clothes and shoes during the measurementprocess.
To verify the results the second sample was taken in the same industrial companies located throughout Serbia in 2013. The second sample size is 50 participants. All the participants were male, with an average age of 46, and standard deviation of 11.8 years. The same procedure was applied and the results are shown in Table 2.
Multivariate modeling on the samples of Serbian metal industry workers
Factor analysis usage for investigating the suitability of the collected anthropometric data is a tool that has not been widely used [29], although the multivariate method which could be successful in overcoming the shortcomings of the percentile method is PCA. It can be used to determine critical anthropometric models that provide the appropriate fitting of the desired percentage of the population [30, 31]. PCA reduces the dimensionality of the data set consisting of a large number of correlated variables, while retaining the variation present in the data to the greatest degree possible. PCA does so by transforming the original data set to a new set of variables, called principal components (PCs), which are uncorrelated and ordered, so that the first few retain most of the variation present in all of the original variables. PCs tolerance ellipse or ellipsoid can be fitted so that for the specified probability of the given contour it contains the desired percentage for the accommodation of the individuals in the population. Afterwards, critical models are found as points on the surface of the ellipsoid [22]. The results of PCA application on the Serbian metal industry workers’ data are given in Tables 3–5.
Table 3 shows the eigenvalues and proportion of variation that is explained by each PC. For the decision on how many PCs to retain several different criteria are given in the literature [32]. The desired percentage of total variation that the selected PCs contribute could be the criterion used to select the number of PCs to retain [32]. Another method that is widely used in PCA is the screening method. The point at which the curve first begins to straighten out is considered to indicate the maximum number of PCs to retain [32]. Taking into account both criteria, in the present case the decision was made to retain three PCs.
Table 5 shows the factor coordinates of the variables or factor loadings which represent the correlations between the variables and the PCs. In practice, the first PC almost always has all positive or all negative coefficients for all variables [32] and in our case it reflects the overall ‘size’ of the individuals. Subsequent PCs contrast some of the measurements with others and are interpreted as defining certain aspects of the ‘shape’ [32]. Table 5 shows that the first PC loads relatively heavily on all variables with a negative sign, with the coefficient for foot length, shoulder width and hip width somewhat lower. Those three variables can be interpreted as describing the ‘shape’ of the individuals and are explained by the second and third PC, as the second one loads heavily only on shoulder and hip width and the third only on foot length. Accordingly, PC1, which accounts for 45.7% of the total variation, can be interpreted as the workplace space ‘height’. PC2, accounting for 21.2% of the variation, calculated mostly from shoulder and hip width, is interpreted as the workplace space ‘width’ and PC3, accounting for 9.4% of the variation, which mostly explains variable foot length, is interpreted as the workplace space‘depth’.
In this study, we used the tolerance ellipsoid rather than prediction as proposed in [33], with 95% of the desired percentage of the population to be fitted, with 95% confidence. The obtained ellipsoid is given in Fig. 1 along with the critical models. There are 14 points representing the critical models on the surface of the ellipsoid, 6 of them at the intersection of the axes and ellipsoid, and the remaining 8 at the centers of the octants.
The factor coordinates of the 14 critical models from the accommodation ellipse are transformed back into 8-dimensional anthropometric data by multiplying the matrix of the factor scores with the inverse eigenvector matrix. Table 6 shows the resulting matrix, e.g. 14 representative body models with their standardized values for 8 anthropometric measurements. Table 7 shows the anthropometric measurements for those 14 models obtained afterwards by multiplying the standardized values with the standard deviations and adding the score to the mean of the appropriate dimension. The following 14 body models are the representative body sizes and types for Serbia’s metal industry workerpopulation: Model V represents an individual with small overall height, and average width and foot length. Model U represents an individual with large overall height, and average width and foot length. Model Z represents an individual with large width, and average overall height and foot length. Model X represents an individual with small width, and average overall height and foot length. Model W represents an individual with large foot length, and average overall height and width. Model Y represents an individual with small foot length, and average overall height and width. Model G represents an individual with relatively large width and overall height, but small foot length. Model A, in contrast to Model G, represents an individual with relatively small width and overall height, but large foot length. Model D represents an individual with relatively small overall height, but large width and large foot length. Model F, in contrast to Model D, represents an individual with relatively large overall height, but small width and small foot length. Model E represents an individual with small width, but relatively large overall height and foot length. Model C, in contrast to Model E, represents an individual with large width, but relatively small overall height and foot length. Model B represents an overall small individual. Model H, in contrast to Model B, represents an overall large individual.
A graphical representation of two of these models is given in Fig. 2, in both standing and sitting positions.
Figure 2 presents two critical models from the surface of the ellipsoid; model G and model A. Their positions on the ellipsoid are opposite to one another, which implies the contrast in their measurements.
We verified the presented method with a control sample consisting of 50 participants. The factor coordinates of these 50 participants were calculated using the same procedure previously described. Those coordinates were placed in the first sample’s enclosure space and in that way we confirmed that 96% of the participants in the control sample are within the ellipsoid.
Comparing PCA and percentile method results
In order to compare the percentile method and the PCA method, we identified the minimum and maximum anthropometric measurement values from the 14 critical models, and their corresponding calculated percentiles. The 5th and 95th percentiles of these dimensions are given in Table 8.
The traditionally used percentile method implies that if individuals with anthropometric measurements in the 5th to 95th percentile range are accommodated, then 90% of the population will be accommodated. A more stringent requirement specifies the accommodation of individuals with anthropometric measurements ranging from the 3rd to the 98th percentiles, which implies that 95% of the population will be accommodated. However, when design includes more than one anthropometric measurement, the actual percentage of individuals accommodated is quite lower. For instance, taking our sample into consideration, if we eliminate those individuals who are not included within the 5th to 95th percentile limits (Table 7) on any of the 8 measurements, 43 persons will be excluded, amounting to 35%, which is far more than the anticipated 10%.
Another issue with the percentile method is that it does not take into account individuals with extreme combinations of measurements [18]. Figure 2 shows that models A and G both have extreme combinations of measurements. Model G (left) is of relatively tall stature (1840 mm, which is equal to the 70th percentile), and has wide shoulders and hips (565 and 468 mm, i.e. the 90th and 89th percentiles) but relatively small foot length (260 mm, i.e. the 20th percentile). In contrast to Model G, Model A is of relatively short stature (1732 mm, i.e. the 20th percentile) and has narrow shoulders and hips (377 and 335 mm, i.e. the 2nd and 1st percentiles), but relatively long foot length (273 mm, i.e. the 70thpercentile).
Conclusions
Occupation is known to be a factor which influences the variability of anthropometry data, and since the metal industry represents one of the most important Serbian economic sectors, the subjects of this paper are metal industry workers.
When each dimension is sequentially arranged to cover a certain percentile population, the design will include a certain percentage of the user population for each specific function, but will suffer from a compounded decrease in the level of overall accommodation, which in turn will result in workspace determination inefficiency. This study, which relies on three PCs, shows that linear combinations of the 8 original variables are very useful in addition to the percentile approach. The models we generated in this study include not only overall large and small persons, but also individuals of different body configurations. We verified this method with a control sample and it shows that 96% of the factor coordinates of the control sample are within the enclosure ellipsoid. Hence, industrial workplaces designers can use these models together with the percentile approach to determine or verify their designs. We propose method application on a larger sample that will include female industrial workers, who are less represented inSerbia, and this can be seen as the limitation of this study. It would also be interesting to compare the results obtained here to those obtained using the method of stepwise combination of pairs of data, where each combined pair became an input for the succeeding step, as proposed in [34].
A number of problems, such as how to minimize or remove the anthropometric mismatch and improve work posture thus minimizing the risk of musculoskeletal disorders by applying the combined approach proposed here could be resolved using the modeling approaches shown in this paper when designing metal industry workspace. After eliminating the anthropometric mismatch, the increased enthusiasm of workers, a reduction in work-related pain and injury risk, an increase in the comfort and efficiency of operators, and an increase in overall system productivity and safety could be expected.
Further analysis of different working professions and the effect of the anthropometric information obtained in this survey and their exact accommodation modeling is expected in further studies. Also, comparing different national industrial anthropometric data in European countries will be useful in finding the common denominators important for the manufacturers of metal industry machinery andequipment.
Conflict of interest
The authors have no conflict of interest to report.