Final Report

NASA NAG 5-3990

Principal Investigators:

Norman I. Badler, University of Pennsylvania

Dimitris N. Metaxas, University of Pennsylvania

Dava J. Newman, Massachusetts Institute of Technology

August 24, 2000

Executive Summary

Our NRA NAG 5-3990project had three overall goals. The first was to investigate dynamic simulation techniques tailored to microgravity IVA and EVA activities. The second project goal was to produce and evaluate a human performance model in a realistic NASA mission situation. The third goal was to develop a representation and software infrastructure to support the simulation of virtual autonomous crewmembers for training and task evaluation. This technique would be used to measure and evaluate task feasibility as well as develop new methods of task assignment in space related activities. All of these projects would enhance ground-based IVA and EVA procedure design as well as predict payload-handling difficulties. Moreover, the potential predictive power of workload models will enhance our computational understanding of other Earth-bound activities. We developed an efficient human motion planning method based on recursive dynamics and optimal control techniques, which are subjected to minimum torque criteria. We also performed workload analysis (metabolic load, stamina, and fatigue). Applying biomechanical modeling and physics-based dynamic simulation can establish analytic and predictive measures for IVA and EVA space human factors. This project extended existing dynamic, anthropometric, and kinematic models for human motion.

Outcomes

Our three project thrusts on NRA NAG 5-3990 were:

  • To investigate dynamic simulation techniques tailored to microgravity IVA and EVA activities.
  • To produce and evaluate a human performance model in a realistic NASA mission situation.
  • To develop a representation and software infrastructure to support the simulation of virtual autonomous crewmembers for training and task evaluation.

The first item is reported in Appendix 1 as a paper presented at the 1999 US-Japan Space Human Factors meeting. The second item is reported in the discussion below, as it has the most direct bearing on NASA mission issues. Two papers developed from this task are included as Appendices 2 and 3. The third item is also summarized in the discussion below, and elaborated in Appendix 4.

An Evaluation of Human Performance using Dynamics Models

Astronaut performance, specifically extravehicular activity (EVA) tasks, were investigated and modeled numerically. Contributions included human factors insights that come through modeling as well as a physics-based astronaut model and a mathematical representation for the extravehicular mobility unit (EMU), or space suit. Computational multi-body dynamics were used to simulate astronaut extravehicular activity (EVA) tasks. Two actual EVAs were simulated: manipulation of the Spartan astrophysics payload on STS-63 (large mass handling) and attempts at capturing a spinning Intelsat VI satellite on STS-49. This research effort fills a current gap in quantitative analysis of EVA by employing computational dynamics, with emphasis on Kane’s method, to solve the equations of motion for the dynamics of the astronaut’s body segments and other interacting objects. The simulation approach can be divided into six phases: (1) model design, (2) system description, (3) equation formulation, (4) inverse kinematics, (5) inverse dynamics, and (6) data display with animation. The large mass handling simulation was performed using a relatively simple seven segment astronaut body model with 6 degrees of freedom and motion restricted to a single plane. Results of the modeling effort reveal how an analyst might predict difficulties imposed by task specifications requiring violation of physiological limits, and modify the protocol so that the tasks objectives are humanly achievable. The more complex Intelsat EVA investigation, using a 12 segment astronaut body model with 31 degrees of freedom, and interacting capture bar and satellite objects, each with 6 degrees of freedom, reveals greater challenges in terms of motion control and numerical integration. Interaction between the capture bar and satellite is modeled by means of constraint forces imposed at two contact points and achieves realistic motion of the two objects. This work has resulted in a recent publication [Sch00]. Another contribution of this research effort was to develop a dynamic model of the extravehicular mobility unit (EMU), or current NASA space suit. The EMU model incorporates three key suit parameters, namely, mass, inertia and performance for EMU components including the portable life support system (PLSS). Replicating the Spartan EVA simulations while including the space suit model reveals that the astronaut does nearly an order of magnitude more work to produce the same results when the initial conditions are such that the lower body is fixed. Allowing for a compliant lower body astronaut model with the suit model results in the suited and unsuited condition requiring similar amounts of work by the astronaut. An interesting result is that the initial conditions from which the astronaut starts the task greatly affects the results (i.e., astronaut neutral body posture versus EMU neutral suit posture in microgravity). Finally, a series of simulations was performed to assess the effect of a space suit on an astronaut engaged in repetitive motions over a long time representative of future International Space Station (ISS) tasks. The accumulation of additional work to overcome space suit properties might lead to accelerated muscle fatigue during these simulations. This work was most recently published and presented20 at the International Conference on Environmental Systems (ICES 2000).

Background

Motivation

This work fills a current gap in quantitative analysis of EVA by solving the equations of motion for the manipulated objects and a multisegment human model. The application of computational dynamic simulation to EVA was prompted by the realization that physical microgravity simulators have inherent limitations: viscosity in neutral buoyancy tanks; friction in air bearing floors; short duration for parabolic aircraft; and inertia and friction in suspension systems. Existing useful computer programs either produce high resolution three-dimensional computer images based on anthropometric representations [Pri94, Bad93], or empirically derived predictions of astronaut strength based on lean body mass and the position and velocity of the body joints [Pan92], but none provide dynamic analysis of EVA tasks using the equations of motion of a multibody system model.

Several classical methods of formulating the equations of motion of a multi-segment system exist. These include the methods of Newton-Euler, Lagrange, D'Alembert, Hamilton, Boltzmann-Hamel, Gibbs-Appell, and Armstrong [Hoo65, Rob66, Wit79, Sil82]. In fact, one of the early applications of multi-segment dynamic analysis to human motion involved astronaut body orientation in weightlessness [Sch69]. Since this early study was performed in 1962, computational methods had not yet been developed and it took the analysts weeks to derive the equations of motion by hand. Other novel methods have been developed more recently [Ram80, Fea83, Arm85]. Even more recently, some researchers have started to explore the power of computer graphics and animation in multi-segment dynamic system analysis [Wil88, HODGINS????? Hua00, Lo, Hua].

A particularly efficient multibody dynamics equation formulation method has been developed by Kane and his associates [Kan83a, Kan83b, Kan83c, Sch88]. An outgrowth of their work was the development of highly efficient algorithms for multibody analysis [Ros86] that have been incorporated in the computer program SD/FAST [Hol94]. The simulations discussed herein use SD/FAST to formulate the equations of motion for the dynamic system being modeled.

Eventually, the goal of our dynamic simulation effort is to analyze EVA tasks before they are performed in space to obtain numerical estimates of expected dynamics parameters, such as astronaut joint angular excursions and torque requirements, and to identify possible difficulties that can be further examined using physical simulators. In this stage of development, it is advantageous to compare simulation results with the actual EVA as a means of validating the computational simulation program. Unfortunately, there are few, if any, means of obtaining dynamically relevant numerical data on astronauts performing EVA in microgravity, therefore, it is not yet possible to make direct comparisons of dynamics parameters.

The result of the first year’s research effort was a 7 segment astronaut model with a limitation that the astronaut model did not account for the mechanical influence of the space suit on performance. The second year’s research effort has resulted in an enhanced three dimensional 12 segment astronaut model and a dynamic space suit model described in detail in following sections.

Research Objectives

Certain specific objectives were established to guide the research effort and are listed below:

  • Develop a convenient means of modeling the dynamics of an EVA astronaut.
  • Transform the description of the dynamic system into equations of motion represented in computational form.
  • Develop computer code to drive simulations of the dynamic system under a variety of conditions.
  • Explore methods of prescribing the motions to be performed in a task-oriented form, the way that an astronaut or trainer might think of the operation, without the need to explicitly specify the kinematics (positions, velocities, and accelerations) of each segment. In other words, perform an inverse kinematics analysis, given only the motion of the endpoint of the system.
  • Determine the joint torques required to drive the system in performing a particular motion by using the calculated segment kinematics in an inverse dynamics analysis.
  • Enhance the 7 segment astronaut model to provide realistic motion using a 12 segment model.
  • Develop a dynamics model of the EMU in conjunction with the enhanced astronaut model.

Methods

The dynamic simulation approach used in this study may be divided into six phases: (1) model design, (2) system description, (3) equation formulation, (4) inverse kinematics, (5) inverse dynamics, and (6) data display with animation.

In the first phase, model design, the analyst develops a conceptual model that represents the dynamic system to be analyzed in sufficient detail to ensure that the desired accuracy is obtained. This includes determination of the system geometry (number of segments, their dimensions, how they are linked, and the degrees of freedom of the joints linking the segments) and the mass properties of each segment (mass, moments of inertia, and products of inertia when necessary).

The second phase, system description, involves writing a system description file. Code words and numerical values that fully describe the relevant geometry and mass properties of the dynamic system are included in this file in a format that can be interpreted by a computer program capable of formulating the equations of motion of the system (we use SD/FAST).

In the third phase, equation formulation, the formulation program is executed using the system description file as input. The output is a set of functions (subroutines) in C code that implicitly represent the equations of motion of the system and aid in the analysis of the dynamics of the system.

The fourth phase, inverse dynamics, combines the equation formulation code with user-written code that performs the actual simulation run. Our code does this operation in two parts. The first part prescribes the motion of a few selected segments (usually the astronaut’s hands), representing a ‘task’, and solves for the motion of the remaining segments subject to any external or internal loads (forces and torques). If there are more degrees of freedom than constraints in the system, the solution is found using a linearized least squares solver. The second part prescribes the motion of the entire system, using the results from the first part, and then calculates the joint torques required to achieve the prescribed motion.

The fifth and sixth phases, data analysis and animation, involves interpretation of the results using graphical methods. Simulation data, usually position and torque time histories, are plotted on two-dimensional graphs and then evaluated for range of motion or strength extremes, efficiency and comfort of task performance [Hua99]. Three-dimensional rendered animation is a powerful tool in evaluation of results. It allows the analyst to quickly determine whether the starting configuration and subsequent motion fall within reason. It is also extremely helpful in determining the cause of errors or anomalies in the simulation.

In creating the initial model of a dynamic system, it is advisable that the analyst start with the simplest possible model capable of exhibiting the required dynamic characteristics with a reasonable degree of accuracy. Once a working model has been obtained, the complexity (number of bodies and degrees of freedom) can be expanded incrementally to study other effects or increase realism. Simulations to illustrate the modeling efforts are presented. The relatively simple seven-segment “astronaut” model with an additional eighth segment representing a large payload being manipulated (Spartan astronomy satellite) was an initial step. The enhanced twelve-segment astronaut model with a total of 31 degrees of freedom is modeled representing full three-dimensional movement capability. The astronaut interacts with two additional segments, each with 6 degrees of freedom, bringing the total number of degrees of freedom to 43. Also, a dynamic representation of the EMU is modeled 43.

A Dynamic Model of the EMU

A data-driven dynamic model of the extravehicular mobility unit (EMU) has been constructed based on three key suit parameters: mass, inertia, and performance. Mass data for the individual EMU components, including the PLSS, were obtained from Hamilton Standard mass properties reports. Various moments of inertia were then computed based on suit dimensions, and entered as parameters in a system description file in the SD/FAST dynamic simulation environment. Similar to existing files created for previous simulations, these system descriptions were then used to generate the (passive) equations of motion for the various suit segments. These suit segments are then attached directly to our new 12-segment human model, delivering the inertial loads of the suit to the astronaut’s limbs.

Suit-Imposed Torques

The most profound effect of the EMU on astronaut performance involves the imposed torques, or “springback forces,” in the joints; these forces are generated when the suit fails to maintain a constant volume during movement, and the astronaut does work to change the suit position. Also, the multi-layer soft shell construction of the suit behaves much like a giant winter coat: as the limbs move back and forth between joint limits, the suit fabric tends to bunch up, and eventually the sheer volume of material compressed into a small volume creates a firm limit on the maximum flexibility of a joint.

EMU design, of course, attempts to minimize these effects in order to maximize suit flexibility and astronaut mobility. For the most part, suit designers have succeeded; joint restraints are quite effective in maintaining a nearly constant joint volume through the most frequently used areas of an astronaut’s workspace, and innovative designs such as flat-patterned mobility joints allow easy flexing of joints without substantial twisting, bunching, or stretching of thermal micrometeorite garment (TMG) material. These joints are basically folded pleats or tucks in the outer layer of the TMG which allow layers of the suit to fold over themselves in an orderly fashion, rather than crumpling in a less efficient way.

The effects being modeled, then, are not the major suit forces associated with early space suit designs that lacked these improvements. Rather, we attempt to model the EMU’s deviation from an “ideal” suit, one, which would exert no forces to counteract astronaut movement. In practice, subtle changes in the orientation of joint restraints during a particular motion can lead to minute variations in the restraint forces applied to EMU joints. Additionally, regardless of exceptional joint designs, TMG fabric is thick and bulky, and it will contribute some small countertorque to astronaut movement. These variations lead to slightly varying joint volumes during particular sequences of arm and leg motion, and increasing fabric bunching toward the joint limits; the result is the characteristic hysteresis joint angle vs. suit torque curve of Figure 3, obtained from torque measurements of a pressurized suit elbow over its range of motion. NASA personnel at both JSC and Ames Research Center have been instrumental in providing this data.

It is clear from the description above that an analytical model for the “error” in slightly non-optimal suit joints would be extremely difficult, if not impossible to derive from physical principles. This modeling effort has therefore focused on developing a dynamic model of the EMU from existing suit, which presumably represents some combination of the effects that various suit imperfections have on suit performance. Although mass and inertial properties provide a more complete model, these springback torques represent the critical element for assessing detriments to human performance in the EMU.

Hysteresis model

The hysteresis space suit model results in the torque experienced by a suited astronaut performing representative tasks. The plot exhibits a significant degree of hysteresis, revealing that suit-applied torques are dependent on the direction of arm motion. As with any hysteretic system, this means that the magnitude of the torque applied by the suit at any moment in time depends uniquely on the particular history of arm movements up to that point. The soft EMU joints have a “memory,” storing the sequence of bunchings and expansions that work the TMG into a particular orientation and energy state; that unique state then dictates the amount of energy required to move the joint to another position.