ERGONOMICS IN DIGITAL ENVIRONMENTS:HUMAN PERFORMANCE MODELS

HUMAN PERFORMANCE MODELS

Human capability analysis is one of the primary motivations for simulation. Commercial modelers have implemented performance models from the academic literature into their software, taking ad- vantage of the human figure sophistication and real-time visualization technologies. A review of the commonly available performance models reveals that independent research groups largely developed them. The development diversity is reflected in the variety of inputs that these tools require in order to provide an ergonomic assessment. This represents a challenge to the modelers as they work to seamlessly integrate these assessment tools into their animation and simulation offerings. Some tools lend themselves well to integration, such as those that can capture all required information from posture, figure, and load mass characteristics. Typically these are the tools that have as their foun- dation biomechanical models from which the inputs are derived. Others, which were originally in- tended to be used with a checklist approach, are more challenging in that they often require complex questions to be answered that are straightforward for a trained ergonomist but quite complex for a computer simulation system (Table 2).

Most often, simulation engineers expect to ask human performance questions of their simulation without having to redescribe the simulation in the language of the tool. Thus, ideally, the tools are implemented such that they can derive all the necessary information from the simulation directly. Less ideally, the engineer performing the assessment must explicitly identify details of the simulation using tool specific descriptors.

Strength

Strength assessments are a typical human performance analysis, regardless of whether the application involves manual handling tasks, serviceability investigations, or product operation. Questions of strength can be posed in a variety of ways. Designers may want to know the maximum operating force for a lever, dial, or wheel such that their target demographic will have the strength to operate it. Or the engineer may create a job design and might ask what percentage of the population would be expected to have the strength to perform the required tasks of the job. Strength data might also be used to posture virtual human figures by using an algorithm that optimally adjusts the individual joints of the manikin to produce most effectively the required forces for a task.

A large amount of strength data has been collected over the past quarter century. Most of these have been in the form of maximal voluntary exertions (MVEs) of large muscle groups. Subjects are placed in special strength testing devices (e.g., strength chairs) to isolate individual muscle groups, and standard methods controlling for repeatability and duration of effort are then used to capture the

a The tools require different types of input that often cannot be accurately deduced from an animation sequence, requiring tool specific user input.

strength levels accurately. Strength can also be assessed while taking into account a subject’s per- ception of the effort. These data, in which subjects’ impression of the load strain is included, are called psychophysical strength data. They differ from the maximal voluntary exertions in that they are more task specific. Subjects are asked to identify the load amount they would be comfortable working with over the duration of a work shift. Typically, these are laboratory studies in which mockups of tasks very close to the actual work conditions are created and subjects are given an object to manipulate to which weight can be added. The worker has no knowledge of the weight amount (bags of lead shot in false bottoms are often used), and experimental techniques are employed to converge on the weight that the subject feels comfortable manipulating over the course of a workday. The data of Ciriello and Snook (1991) are of this nature. Finally, a few dynamic strength data have been collected. These data are complex to collect and for this reason are also the most scarce. Specific dynamic strength-testing equipment is required to control for the many variables that affect dynamic strength, including the movement velocity and posture. As a consequence of the data- collection limitations, these data are also quite restricted in terms of the range of conditions in which they can be applied, such as the prediction of dynamic strength capability for ratchet wrench push and pull operations (Pandya et al. 1992). Lately, the rise of cumulative trauma injuries for the lower arm, wrist, and hand has created a need for strength data specifically pertaining to the hand and fingers and estimates of hand fatigue. A extensive amount of data is available on grip strengths in various grip postures, but these data, because they do not adequately describe the distribution of exertion loads on the individual digits, do not lend themselves well to the estimation of hand fatigue issues. This problem is compounded by the challenge of accurately posturing the hands in digital models. There are many bones and joints that allow the complex movement of the fingers, most of which are included in contemporary human models. For example, the Jack human model has a total of 69 segments and 135 DOF, of which 32 segments and 40 DOF are in the hands alone. While a solution to the manipulation of these many degrees of freedom is presented in the section describing motion-tracking technologies, models that are able to analyze the postures and gripping conditions are still needed before hand fatigue issues can be addressed.

Whole body strength data in contemporary human models are available in a range of forms, from simple data lookup tables to statistical equations that are used in conjunction with biomechanical models to drive a measure of population strength capability. In the United States, perhaps the most widely used strength data are from the University of Michigan Center for Ergonomics (3DSSPP) and Liberty Mutual Insurance Co. (Ciriello and Snook 1991). Within the defense industry, the CrewChief strength data are also popular because the modeled strengths were obtained from military-related maintenance tasks. In Europe, particularly Germany, the data of Burandt and Schultetus are often used. As mentioned previously, a few of these data were obtained without the intent to incorporate them into human models. Instead, the data are presented in tables indexed by such factors as loading condition and gender. Those data that were collected and described with a focus toward eventual human model inclusion tend to be formulated such that all relevant information needed for a strength assessment can be deduced from the human model mass, loading, and posture information. These strength models now are very attractive to the human modeling community because they afford the real-time assessment of strength issues during a simulation without the user having to identify data- specific parameters or conditions (Table 2).

As discussed above, the availability of dynamic strength data is limited. The narrow scope of applications to which these data can be applied restricts their attractiveness to human modelers and the user community. An interesting associated note regarding these dynamic data and human models is that even if these data were more complete, the difficulty in accurately determining movement velocities from simulations would affect their use. Unless the virtual human motions are defined via motion-capture technology, the designer’s guess of the movement speeds is fraught with error. Even under conditions in which actual motion capture data are used to animate the virtual figures, the velocities are derived by differentiation of the position information, which can result in noisy and unreliable input to the dynamic strength predictors. However, because people undeniably move during work and dynamic strength capability can differ greatly from static, this is clearly an area that will likely see research and technological attention in the near future.

Fatigue / Metabolic Energy Requirements

Once a simulation of a virtual human performing a task has been created, questions regarding the fatigue of the worker are commonplace. Can the worker be expected to perform at this cycle rate, or do we have to decrease the line rate or otherwise reengineer the task to avoid worker fatigue? Research suggests that whole-body fatigue increases the risk of musculoskeletal injury through pre- mature decrease in strength. Unfortunately, the available empirical data are largely inadequate to predict a worker’s fatigue level accurately. The lack of data can be explained by the large number of variables that affect fatigue, including exertion level, dynamism of the exertion, muscle tempera- ture, previous exertion levels, the muscle groups involved, and individual conditioning. Nevertheless,

two approaches are currently in modeling practice to provide at least some level of quantitative fatigue assessment for a work task. The strongest of these from a data perspective is the use of empirical metabolic energy prediction equations, in particular the equations published by Garg et al. (1978). These equations model a series of typical industrial materials-handling motions, such as walking, lifting, carrying, reaching, and arm work. Based on motion-specific parameters and load amount, the mathematical models provide an estimate of the energy consumption in kcal / min. These data were validated on a broad range of manual handling activities and were shown to predict actual energy- consumption rates well. The energy-consumption rate can be compared with accepted standards for exertion levels, such as the NIOSH 1991 recommended limits. The guideline put forth in the devel- opment of the NIOSH 1991 lifting equation recommends a limit of 33% of the maximum aerobic capacity of 9.5 kcal / min for healthy individuals performing whole body lifts over an eight-hour day (3.1 kcal / min). For work that involves mostly the arms, NIOSH recommends a reduction of 30% from this level or approximately 2.1 kcal / min (Waters et al. 1993). If the simulated task is found to require a higher energy-consumption rate than the recommended limit, it is assumed that the task is fatiguing and must be modified.

One of the challenges for modelers in using these energy-expenditure models is in the ability to deduce automatically which equations apply to the particular motion under simulation and then to provide the appropriate equation parameters. Some models include these data as a separate tool wherein the user explicitly defines the simulation in terms of the motion sets defined and modeled by Garg et al. A criticism of the approach regardless of implementation is that the granularity of the analysis is large, making it difficult to identify the particular body area that is fatigued, and that the data do not provide information on a broad enough range of activities.

In contrast to this approach, a variety of endurance equations may be used to estimate the amount of time static exertions can be held (see Van Diee¨n and Vrielink 1994). These equations describe the amount of time subjects can perform static exertions at various levels of effort relative to their maximum capability. Relevant to industrial work, some of these include the effects of interspersed rest periods (Sjogaard et al. 1988). Equations to describe the amount of time required to recover from these exertions were published by Rohmert (1973a, b) and Laurig (1973). If the estimated amount of time needed to recover from an exertion exceeds the amount of time available during a job cycle, then fatigue is assumed to accumulate. The endurance relations are applied to each body area separately, requiring an estimate of exertion level, or percentage of maximum capability, at these areas. While the original subjects were strength tested to derive their strength capability, these data are not available for workers in general and an estimate of strength capability must be used. One solution is to use biomechanically based strength models. A task simulation is analyzed with regard to the postures adopted by the virtual worker, and an estimate is given to the amount of time the worker spends in each of the postural conditions. The level of exertion required is estimated utilizing the strength equations, and this information is input to the endurance equations to provide the recovery time estimate.

While the methodologies for using these endurance data within the modeling tools have been implemented and are in use, the endurance data themselves are limited, as mentioned earlier. Gender and age effects are not taken into account, nor are most of the multitude of other factors that influence fatigue. Only the exertion level under static conditions is considered. However, the need to predict quantitative assessments of worker fatigue in simulations is high enough that users of human models look for ways to obtain a metric of fatigue, working around the limitations of the foundation data. Toward this end, joint use of the energy expenditure equations, endurance equations, and stress analysis using the strength tools will currently provide the best estimate of the task injury potential from fatigue.

Low-Back Injury Risk

Low-back injury is estimated to cost the U.S. industry tens of billions annually through compensation claims, lost workdays, reduced productivity, and retraining needs (NIOSH 1997; Cats-Baril and Fry- moyer 1991; Frymoyer et al. 1983). Approximately 33% of all workers’ compensation costs are for musculoskeletal disorders. Experience has shown that these injuries can be avoided with the proper ergonomic intervention. Biomechanical models available can be used for job analysis either proac- tively, during the design phase, or reactively in response to injury incidence, to help identify the injurious situations. The most common types of injury-assessment analyses performed using human models include low-back compression force analysis and strength analysis.

Low-back pain has been well researched over the past 20 years, including epidemiological studies that have identified spinal compression force as one of the significant predictors of low-back injury. In response, sophisticated biomechanical models have been developed to estimate this compression force accurately, taking into account not only the weight of the object and the body segments but also internal forces generated by the musculature and connective tissues as they balance the external loads (e.g., Nussbaum et al. 1997; Raschke et al. 1996; Van Diee¨n 1997). These internal contributions

to the spinal forces can be an order of magnitude larger than the applied loads. NIOSH has recom- mended guidelines against which the predicted compression forces can be compared and job-design decisions can be made.

Comfort

Assessment of worker comfort using digital models can be based on both posture and performance model analysis. However, since comfort is influenced by a wide variety of interacting factors, these tools are largely insufficient to quantify the perception of comfort with accuracy. Most comfort studies performed to date have been centered around a specific task, such as VDT operation or vehicle driving (Rebiffe´ 1966; Grandjean 1980; Porter and Gyi 1998; Krist 1994; Dreyfuss 1993). Within the boundaries of these tasks, subjects are observed in different postures and asked to report on their comfort via a questionnaire. The joint angles are measured and correlated with the comfort rating to arrive at a postural comfort metric. Because these data are mostly collected under specific, often seated, task conditions, some caution is required to apply these to the analysis of comfort in standing postures such as materials-handling operations. In addition to the posture-based comfort assessment, a variety of the performance tools can be used to help with the assessment of comfort, including strength capability, fatigue, and posture duration information. Certainly the multifactorial nature of the comfort assessment makes it challenging, and perhaps for this reason it is seldomly used in the analysis of physical tasks.

Motion Timing

A typical simulation question regards the estimated time it will require a person to perform a task. Digital human models can draw on a wealth of predetermined time data available. Motion-timing data are collected in studies where elemental motions (walk, turn, reach, grasp, etc.) are observed performed by skilled operators in the workplace, and timed using a stopwatch. The motion time data are then published in an easily indexed form with associated movement codes. The best known of these is the methods time measurement (MTM-1) system published by the MTM Association. This system has the advantage that it has a large number of elemental motions defined, allowing for a precise partitioning of the work motions within a job task and subsequent accurate assessment of the movement time. One drawback of this high resolution is that considerable overhead is required to break the motion into the elemental movements. To address this, the MTM association as well as other groups have published grosser movement times, which combine several elemental movements into one. Several of the human modeling solutions now provide simulation solutions that can define movement duration with input from these movement time systems.