1 Introduction

It is well known that astronauts play a very important role in manned space missions. In order to ensure the successful implementation of manned space missions, it is essential to maintain and improve the capabilities and task performance of astronaut in space [1]. However, in spaceflight, astronauts are exposed to numerous stressors, such as microgravity, confinement, and radiation, all of which may impair human capabilities, and increase task risks [2]. So the mismatch between crew capabilities and task demands has been listed as a major risk that astronauts may encounter in space in the human research roadmap published by the National Aeronautics and Space Administration (NASA) [3]. If tasks prove too difficult for the operators, whether as a result of inadequate design of man-machine interfaces or task schedules, or capability decline in spaceflight, the work efficiency of the space crew may decrease and the likelihood of mission failure increases. So it is crucial to get a better understanding of astronauts’ capabilities and their task performance during spaceflights.

Computational simulations, especially those combined with human performance models (HPMs), have been used more and more frequently in space explorations and aviation to analyze human performance and identify potential human-system errors, due to the cost and efficiency advantages they have over traditional human-in-the-loop tests [4, 5]. HPM simulations can be used early in the development process of a product or system, thus are highly applicable to human-system integration design of space tasks and systems. Specially, some aspects of the spaceflight environmental factors, such as the weightlessness, are difficult to simulate on ground, while HPM may offer an effective mean to simulate human behaviors and performance in such environment, thereby providing useful guidance for human-system integration design.

There are generally two kinds of HPMs that are commonly used in the space and aviation industry: physical models (models of human anthropometry, models of biomechanics), and cognitive models built from empirical research and theories of human cognitive processes; the tasks investigated can also be categorized into two types: physical tasks and cognitive tasks. A lot of software tools have been developed to support human performance modeling. For example, Jack and DELMIA, which consist of models of human anthropometry, have been used to analyze the human reach envelope, human visual field and space interference problem [6]. Anybody, OpenSim and Visual3D, which consist of models of human biomechanics, have been used for force analysis of human body (including joint torque, muscle force, etc.) in physical task [7]. ACT-R and Soar, which consist of human cognitive models, have been used to simulate decision making processes in cognitive tasks [8]. The Man-machine Integration Design and Analysis System (MIDAS), developed by NASA, combines graphical equipment prototyping, a dynamic simulation, and human performance modeling in an integrated platform [9]. Cognitive models are core elements of MIDAS. The aim of MIDAS is to reduce design cycle time, support quantitative predictions of human-system effectiveness, and improve the design of crew stations and their associated operating procedures.

However, most of models and platforms (except MIDAS) are not suitable for the simulation analysis of astronaut performance in spaceflight tasks. They lack systematic considerations of the space environments, astronaut’s characteristics (especially Chinese astronauts’ characteristics in space), and the space task demands, which may all affect astronaut performance in space. As far as we concern, to model Chinese astronauts’ performance in space, especially their performance during long-term spaceflight, cognitive models and physical models should be combined to produce an integrated platform, and a comprehensive model architecture and platform need to be proposed and built.

2 The Architecture and Models of AMSS

2.1 The Overall Model Architecture

The objective of the current paper is to develop a simulation platform to predict human performance in spaceflight. There are three key aspects about the astronaut’s performance modeling: (1) how to use human biomechanical and cognitive parameters to characterize the individual differences, and how these parameters change in spaceflight environmental; (2) how to describe the task features and human-system interaction, such as task type, task procedure, man-machine interfaces, and demands of the task; (3) how to evaluate the human-system effectiveness accurately and quantitatively. Those three aspects were particularly emphasized in our architecture design, and a three-level model architecture, which consists of human characteristic models, behavioral models, and performance evaluation models, were proposed accordingly, as show in Fig. 1.

Fig. 1.
figure 1

Three-level model architecture for astronaut performance modeling

Full size image

In the three-level model architecture, the first level contains a database of astronauts’ characteristics and some models which can describe the changes of astronauts’ capabilities during spaceflight. The second level contains two kinds of behavioral models: biomechanical models and cognitive models. The biomechanical models are used to calculate joint forces, muscle forces, bone stress, bone mineral density in space tasks. Cognitive models are used to simulate the cognitive processes in performing space tasks. On the third level, we build a set of performance evaluation models, so we can predict astronaut’s task completion time, mental workload and physical workload, bone fracture risk, etc.

In this model architecture, the characteristics models mainly reflect the influence of spaceflight environmental on astronaut; the behavioral models mainly reflect the influence of task characteristics on astronauts’ performance. Together with the performance evaluation function, the main human factors concerns are considered in the model architecture.

2.2 Human Characteristics Models

The term “human characteristics” in this paper refers to human biomechanical parameters (such as maximum muscle strength, maximum joint torque), and human perceptual and cognitive characteristics. As many human characteristics will change, it is very important to consider the change regularities of human characteristics in spaceflight when we design a space system. The change regularities of human characteristics in spaceflight are defined as human characteristics models in this paper. The human characteristics models serve as input into behavioral models, and thereby affect the task performance, which reflects the influence of spaceflight environment on human performance. Human characteristics data were collected in spaceflight, or simulated space environment, such as experiments during parabolic flights, the head-down-bed-rest experiment, isolation experiments.

Biomechanical Characteristics

In long duration spaceflight, muscular atrophy and bone loss will have significant impacts on astronaut’s biomechanical characteristics, eventually affect astronaut health and performance. Among the many biomechanical characteristics, the maximum muscle strength, maximum joint torque and the boss density, as well as their changes in spaceflight and simulated weightlessness, are primarily investigated.

The maximum muscle strength are measured in parabolic flight experiments and Shenzhou-10 and Shenzhou-11 missions. As the measurement of these parameters in the parabolic flight experiments and spaceflight missions have many restrictions, such as a short test time in parabolic flights, we also test 16 subjects’ maximum muscle strength, muscle circumference of the shins and thigh, maximum joint torque, bone density in a 45d -6°head-down bed rest experiment. All these data help us to build the models which can reflect the change regularity of the maximum muscle strength, muscle circumference, maximum joint torque [10]. Bone density remodeling is a complicated biological activity, which must be represented in models of nonlinear characteristics. A bone remodeling control equation was used to describe the relationship between density changes and mechanical loads, which can predict bone density in spaceflight [11]. Additionally, the quantitative relationship among bone elastic modulus and bone mineral density was obtained by testing the bone elastic modulus of 3 corpses [12].

Cognitive Characteristics

By cognitive task analysis of the space tasks, a battery of cognitive characteristics that may affect task performance were extracted [12], including perception characteristics such as reaction times, speed perception, time perception, the working memory and prospective memory, spatial ability, fine motor control, attentional control, emotional characteristics, risk decision-making characteristics, etc. Experiments were performed to investigate the influence of spaceflight or simulated spaceflight environmental factors on those cognitive characteristics [14, 15]. Experiments were carried out in China’s Shenzhou-9 to Shenzhou-11 space missions, and in simulated environments such as the head-down-bed-rest-experiments, the parabolic flight experiments, the isolation experiments, and the sleep restriction experiments. Those data were fitted by linear models, and the models were built in the AMSS.

By building up those characteristics models, the effects of spaceflight on human characteristics can be treated approximately in such a way that after the input of the flight state (the astronaut is on the ground, or in orbit, if in orbit, days stayed in orbit, etc.), the models predict the corresponding biomechanical and cognitive parameters, which reflect the astronaut’s capabilities. Those parameters are sent to the behavioral models, and thereby may influence the human task interaction, and the performance outcome.

2.3 Behavioral Models

There are two kinds of behavioral models: biomechanical models and cognitive models. The biomechanical models are used to calculate joint forces, muscle forces, bone stress, bone mineral density in space tasks. Cognitive models are used to simulate the cognitive processes in space tasks. Through these behavioral models, we can describe human system interaction processes and get abundant behavioral data and task performance data. Those data will be further analyzed by the performance evaluation model.

Biomechanical Models

The biomechanical models and the internal relationships of these models are shown in Fig. 2. Because of weightlessness, astronaut’s motion characteristics in space is different from that on ground. Especially, the phenomenon of muscular atrophy and bone remodeling during long-term spaceflight would have serious effect on the performance of astronauts. Biomechanical modeling and dynamic simulation analyses would make it possible to assess astronauts’ movements and predict joint force, muscle force and bone stress. The biomechanical model includes a musculoskeletal system, a kinematics model, a kinetic model, a muscle force prediction model, and a bone stress prediction model. We constructed musculoskeletal system with 72 rigid segments which are corresponding to the anatomy structure of human. Because the shape of muscles is irregular, polygonal lines are used to replace the muscles in musculoskeletal system [16, 17]. The kinematic model takes in the joint coordinates from the musculoskeletal system, and calculates kinematic parameters (e.g. accelerated velocity, angular acceleration of body segment centroid, etc.), which become the input of the kinetic model. Based on Newton’s second law, the kinetic model calculates joint forces and joint torques with input parameters and external forces (as the input parameters of biomechanical model). Similarly, muscle force prediction model calculated muscle forces based on Newton’s second law. We expanded Hill model in calculating the muscle force. The difference between our model and original Hill model is the parameters of muscle lengths and areas are variables which depend on days in orbit. With the muscle forces and joint forces, we calculate bone stress by ANSYS software using the finite element analysis method [18].

Fig. 2.
figure 2

The biomechanical models for simulation of space tasks

Full size image

Cognitive Models

The Adaptive Control of Thought-Rational (ACT-R) has been adopted in cognitive simulations in the platform, as shown in Fig. 3. The cognitive models based on ACT-R for simulation of space tasks. Cognitive processes of specific space tasks are separated into different stages, such as perception, decision-making and manual operation through cognitive task analysis. Covering the stages of cognitive processes, the cognitive model includes goal module, declarative memory module, visual perception module, production system module and manual operation module. The visual perception module serves as the information input of the cognitive model while the manual operation module serves as the information output of the cognitive model. The goal module stores the dynamic goal information of the cognitive model temporarily while the declarative memory module stores the static memory of the cognitive model. The production system module is responsible for central information processing of the cognitive model and interacts with other modules through buffers.

Fig. 3.
figure 3

The cognitive models based on ACT-R for simulation of space tasks

Full size image

In the cognitive model, task-related declarative memory and procedural memory, long-term memory and working memory can be represented, and relevant parameters can be provided to reflect human cognitive abilities and restrictions. Generally, different settings of these parameters will result in different performance outcome in the cognitive simulations [19]. Thus, when the models are validated and the settings of the parameters are accurate, the models have the potential to predict the performance of astronauts in specific space task with different cognitive capabilities, as well as the performance change of the same astronaut when his/her cognitive capabilities changes in the course of spaceflight, which are meaningful for the selection and training of astronauts, as well as the task arrangement in space.

2.4 Performance Evaluation Models

The output of behavioral models are usually raw process data or result data, such as the joint torque, muscle force, bone stress in the whole operation process. To get a good understanding of human performance from those data, performance evaluation models are needed. The performance evaluation models take in the process data or result data, make calculations by specific algorithms, and output more synthetic indices such as the physical workload and the comprehensive performance.

Evaluation of Physical Performance

To evaluate the physical performance, the joint workload evaluation model, fatigue evaluation model, fracture risk assessment model, and muscle strain risk assessment model were built. The relative joint torque, which is defined as the ratio of the current joint torque to the maximum joint torque of the operator, was used as an evaluation index of single joint workload. The upper limb workload evaluation model was established by synthesizing the workload of multiple joints through the method of analytic hierarchy process. In order to analyze the astronaut’s fatigue in specific tasks, the joint fatigue model was established based on the movement time and energy consumption rate. The energy consumption rate is related to the joint torque and the angle of joint movement. Fatigue accumulation and fatigue recovery were considered in the model. Finally, the fracture risk assessment model and the muscle strain risk assessment model were established through the threshold comparison method [20]. The bone stress threshold was obtained through experiments and literature reports [21, 22].

Evaluation of Cognitive Performance

The indices of cognitive performance are rather task specific. In typical space tasks with relatively high cognitive demands, such as the manual rendezvous and docking(manual RVD) task, the control of space manipulator, task complete time, fuel consumption, control accuracy (usually in multiple dimensions) are basic task performance. Comprehensive models synthesizing those multiple indices were proposed and utilized in the performance evaluation in crew training and selection. Moreover, the raw process data of the cognitive models during simulation are sent to a workload evaluation model, which calculate mental workload based on the multiple resource theory [19].

2.5 Model Validation

Models in AMSS has been validated in several ways. The validation method most commonly used is to compare the outputs of the model with the test results of human participants performing the same tasks in spaceflight or simulated spaceflight environment. As some process data, such as the joint torque is hard to measure, we also make comparisons of the outputs of AMSS with some commercial software such as Visual3D, to validate models in AMSS.

Validation of the Biomechanical Models

To validate the biomechanical models, testing equipment and systems were built for biomechanical tests performed on ground, in the neutral buoyance tank, and in parabolic flights. Basic physical operations such as push, pull, lift, press, double arm rotation were performed by participants in the different environment mentioned above, and abundant data such as the body movement, operational force, support reaction force, electromyography (EMG) were collected. The body movement and operational force were input into the biomechanical models, and the models output the support reaction force and muscle activity. The support reaction force that the model output was significantly correlated to that measured in the experiment, the muscle activity that the model output was significantly correlated to the integral EMG that measured in the experiment, which proved that the models have adequate validity.

As some biomechanical data are hard to measure in experiment, we also make comparisons of the outputs of AMSS with some commercial software such as Visual3D, to validate models in AMSS. For the same set of physical tasks, biomechanical data such as the joint torque, the torque angles, the accelerated velocity, the angular speed and the angular acceleration of body segment centroid were simulated by both the biomechanical models in AMSS and the Visual3D, the data output of the two software were highly consistent, which also proved the models’ validity.

Validation of the Cognitive Models

To verify the predictive abilities of the cognitive models, we chose to simulate the manual RVD task, a task commonly required of astronauts during space missions, and compare the performance outcome of the models with that of the human participants.

A research team in China Astronaut Research and Training Center has developed a simulation system of manual RVD by VC ++ language, with the modeling and simulation of the Guidance, Navigation, and Control (GNC) and system dynamics, the docking mechanism, the instrumentation and the TV video. In the manual RVD simulator, the operator, displays, and controllers form a human-in-the-loop system [23]. The operator can observe the information displayed on the monitor and manipulates the controllers to complete the manual RVD tasks. RVD performance, such as control time and fuel consumption, is automatically recorded by the simulation system. The main task processes of manual rendezvous and docking (RVD) include: ascertain the position, attitude, and velocity of the chaser spacecraft relative to the target spacecraft, based on video images and radar information; decide on a strategy for navigating the chaser spacecraft into position; and manually control the joysticks in order to maneuver the chaser into a docking position.

Since the software of manual RVD simulator (developed by VC ++ language) and the software of cognitive models (developed by Common Lisp language) run on different platforms, it is necessary to develop the communication interface between different platforms to enable model-in-the-loop simulations. So a communication interface based on UDP and the multicast techniques was developed, which support data sending and receiving and fulfill the requirement of the real-time communication between the cognitive model and the task simulator.

Comparisons of the task performance including docking precisions, docking processes and fuel cost were made between the cognitive model and skilled operators by correlation analysis. Results showed that the correlation coefficients are above 0.8 (p < 0.01), which demonstrates that the cognitive models could simulate the performance of human participants well [19].

3 The Integrated Simulation Platform

The AMSS has been implemented and integrated through three layers: a user interface layer, a functional implementation layer, and a hardware layer (as shown in Fig. 4).

Fig. 4.
figure 4

The architecture of the integrated simulation platform.

Full size image

The user interface layer facilitates the management and scheduling of simulations and the input for setting task and model parameters. It provides the main interface for entering task-specific parameters and model parameter configurations.

The functional implementation layer enables cognitive and biomechanical simulations, human performance evaluation and analysis, task process visualization, database management and communications among multiple models. The cognitive simulation module is responsible for the simulation of human cognitive processes, while the biomechanics simulation module performs biomechanical analysis of the operator completing specific tasks in space. Task performance analyses are carried out by the performance evaluation module. The 3-D visualization module provides a geometric virtual human, task-specific images and videos necessary to visualize task execution processes [24]. The network communication module controls communication and data sharing among the multiple modules within the platform. The database module, which ties to the hardware layer, stores recorded data from simulations and performs data processing functions such as adding, modifying, deleting, querying, or browsing the data.

Finally, there is a hardware layer, which consists of a computer cluster that provides the simulation platform with high-performance calculation capabilities.

The AMSS has been used to perform the quantitative evaluation of the human-machine interface designs of China’s space lab and the on-going space station missions. In the ergonomic design and evaluation of the display interfaces of the manual RVD system in China’s space lab missions, AMSS was employed to simulate and predict the dynamic mental workload and human performance for operators with different cognitive characteristics and for different interface schemes. Such simulations and analyses provided helpful information for the interface designs. In China’s ongoing space station design, AMSS has been employed to evaluate the workload and fatigue levels of physical tasks, and proper work postures of the astronaut and the strength range requirements are derived from those simulations. AMSS will be used more widely in the ergonomic design and evaluation of China’s space station missions in the coming years.

4 Conclusions

In the current paper, we introduce the Astronaut Modeling and Simulation System (AMSS), which supports modeling and simulation of astronaut’s performance in specific physical and cognitive tasks during spaceflight. AMSS is the first integrated modeling and simulation platform for the human-system integration design of long-term manned space missions in China. A three-level model architecture has been proposed, which consists of the human characteristic models, the behavioral models (cognitive and biomechanical) and the performance evaluation models. Multiple models are integrated in AMSS. The ability to visualize the virtual environment of space vehicle, the virtual astronaut, the operator’s performance and task processes makes AMSS a more user-friendly platform. Models in AMSS has been preliminarily validated by experimental data. AMSS has been used to perform the quantitative evaluation of the human-machine interface designs of China’s space lab and the on-going space station missions.

In the future, more data should be collected during spaceflight or simulated environment to enable better understanding of the impacts of long-term weightlessness, isolation, change of circadian rhythms and other spaceflight related factors on human characteristics. The characteristics models, the behavioral models and the performance evaluation models should be improved and enriched continuously. The software interfaces also need to be improved and updated to better satisfy the demand of the engineers in the space industry.