4.1. General Discussion
The overall findings largely confirm the hypotheses. The assumption of significantly increasing mental workload with high complexity was confirmed, although the factual increases and decreases in the ECG indicators were not particularly pronounced. At the same time, HRV indicators dropped significantly. The increase of choices and corresponding information input not only led to relatively more physical activity and longer assembly times, but also to increased cognitive activity, which in theory contributes to increased sympathetic activity via CNS–ANS integration and its disinhibitory effect. This finding, which has been shown in numerous other research projects [
30,
31,
36], has also received further confirmation and clarification via the eye-tracking indicators. There was an expected increase of fixation duration, which describes the temporal extension of specific information extractions during working on a task. This is also indicated by an additional interaction effect with the assistance system, which is primarily due to the opposite tendencies in the processing of M3 and M4. Only pupil dilation in AR-glasses showed a slight linear increase from M1 to M3 followed by a decrease. Paper and tablet, however, remained at a relatively constant level with no clear trends. This means that the information content and thus the MWL initially increased slightly, but then remained relatively constant at one level in the following models. Subjects have presumably become familiar with the type of task and the associated choice making, search, and assembly activities over the first two models, which reduces the need for additional information extraction. A comparably ambiguous finding can be seen with regard to SPV. There were no significant changes with increasing complexity, whereby interaction effects with the assistance system were evident again.
The examination of AOIs provides additional information on the cognitive activities when subjects concentrate (e.g., on an aspect of the instruction or the object to be assembled). Each gaze activity has the aim to extract information and to initiate manual assembly activities. In this sense, McCarley and Kramer [
43] refer to a close coupling of eye movements and object-related actions that bring the targeted action closer to the goal with the smallest steps. The eye and corresponding hand movements that occur during assembly are not random, but follow very different strategies for integrating information acquisition and behavioral performance [
61]. For example, it seems strategically sensible to first take note of the instructions, then search for parts to be installed, remove them from the container, and finally assemble it, then turn back to the instruction, check the progress, and prepare the next step. The eye movements thus support forward-facing mounting, and backward-looking fixations (in this study revisits to the assistance system) primarily serve as self-confirmations.
The used assistance systems had no independent beneficial or reducing effect on MWL. They do not affect the execution of tasks with low or high complexity. Performance results are solely determined by the quality of the instructions. Thus, it is the content but not the form of presentation that is decisive for efficient working. This finding should not hide the fact that in practice, instruction maintenance is often neglected [
56], probably due to a widespread mindset that assembly work is largely based on routine and cognitive automation. Only when tasks become more demanding and ask for specific voluntary choices during the assembly process–Ballard et al. [
61] speak of so-called what/where modules–should support by assistance systems be offered. Individual interaction effects of the model and the assistance system are presumably due to special features of the AR-glasses, which necessitate a higher adjustment effort.
4.2. Limitations
Cognitive neuroergonomics has two goals: first, to use existing and emerging knowledge of brain function to design technologies and work conditions for safer and more efficient operation, and second, to advance understanding of brain function in relation to real-world tasks and everyday work [
25]. Thus, a major aim of this approach was to validly identify conditions of high cognitive load or mental overload at the real work places and thus to indicate starting points for ergonomic countermeasures. Although our findings support the assumption that more complex tasks lead to higher MWL, we have to realize various inconsistencies in the data. Thus, doubts concerning the view of the MWL construct and its operationalization via different indicators in relation to the fulfillment of natural tasks outside the laboratory have arisen. Many natural tasks used in the laboratory are still close to very narrow experimental paradigms of cognitive psychology [
62], to the simulation of greatly reduced driving situations [
63], or to the detection of situation embedded, salient stimuli on a monitor [
18]. Three central limitations of this study at hand but also of the general approach should be discussed: the relation to the everyday work tasks, the concept of complexity, and MWL measurement using various-related-indicators, leading to the question whether MWL is a unitary or a multidimensional construct [
18,
64].
In the present study, the intention was to reproduce a real-life assembly activity as a natural task in the laboratory. However, associated boundary conditions of an operational assembly activity could not be optimally transferred to a laboratory task. This applies, in particular, to aspects of the qualification and assembly experience of the employees (respectively subjects), but also to the organization of work and commitment to the task. From a cognitive ergonomic point of view, our subjects did not have a sufficient mental model of the assembly work imposed on them [
17]. For example, an attempt was made to operationalize temporal restrictions of real tasks by using a gamification arrangement, but could not actually be enforced stringently.
This missing of effective time limits made subjects rather free to choose their own speed to exert their task. At the same time, this jeopardizes a central assumption of the study. Complexity cannot be conceptualized by solely increasing the number of choices but only in connection with a specified time pressure. Both the number of choices and time pressure are necessary conditions to increase the MWL. If subjects have as much time as they want to select and search parts, complexity disappears. Those who can take any amount of time during assembly do not experience uncertainty or an acute burden. In addition, the lack of experience with assembly activities implies the need to process instructions more intensively, especially at the beginning. This may result in an unexpectedly high MWL during the assembly of M1. After completion of M1, phases of information extraction became shorter due to practice and learning effects over the repeated assembly tasks. This in turn counteracts the assumption that increasing complexity in models 3 and 4 leads to more information extraction and more MWL. Supplementary, for paper- and tablet-based systems, mental models already exist whereas AR-glasses are still new and therefore presumably require more effort and longer familiarization time.
Taking these considerations into account, our operationalization of complexity alone via the number of choices in restricted time showed a further shortcoming. It does not consider the fact that there are dependencies between sequences of choices, which ultimately—and this is shown by the AOI protocols–lead to a different interaction with the information presenting assistance system. From model to model, subjects concentrated their attention more and more on the assembly object. In the course of a concrete assembly, complexity and uncertainty decrease with the reduction of degrees of freedom due to the restricted possibilities to add further parts and give room for more intuitive actions [
65]. In addition, learning processes reduce search times for parts and facilitate the selection of tools. All of these aspects of a natural task contribute to the fact that the complexity of the task was continuously reduced by repetition [
66]. Taken together, natural tasks such as assembly or other manual work should not be seen as static individual action but as a dynamic variable over a period of time. The question arises whether and to what extent natural tasks fulfill the experimental presumptions of a stable and interindividual invariant entity.
A final aspect concerns the measurement of varying MWL that was carried out using ECG and eye-tracking indicators. This combination corresponds to recommendations to not focus on a single indicator solution of an assumed multimodal construct like MWL [
30,
32]. Both groups of indicators we used relate to very different organismic systems which, however, experience central control prefrontal areas of the cortex [
37]. This might support a more unitary view of MWL. In contrast, the group of ECG indicators describes relatively slowly developing changes while the eye related ones showed quicker adaption processes, although both were initialized via the same autonomic nervous system. Thus, there probably is no simultaneously organized activity to be measured. This might favor a view of a more multidimensional approach. Due to improved measurement technology, we are able to measure rather precisely the fixation times or pupil dilations as well as R-peaks and HRV data. However, it is difficult to relate these data to one another for a defined point or period in which mental stress and overload could have occurred. An important prerequisite for a multidimensional measurement of MWL could be to coordinate the spatio-temporal sensitivity of both measures and to show that there is some common variance. Our discrepancies between ECG and eye-tracking data raise doubts about the idea of a unitary concept.
The dimensionality debate is primarily theoretical. For the practical ergonomist, the main question is how to calculate a person’s MWL with regard to a task. For practical purposes, it is not enough to theoretically accept a redline you have to be able to determine how pronounced the MWL is and where there are lower and upper limits that should not be exceeded. There is a lack of possibilities to determine such absolute extents of MWL and normative limits [
10]. For this reason, comparisons between conditions or groups of people predominate in research. This is an important field of work for the future. In a multidimensional perspective, this indicates determining the several limits of different measurement techniques and indicators. At the same time, standards would have to be developed for each measurement procedure, particularly if it is used in the context of natural tasks.
A more practical concern might be the influence of the surrounding on pupil size during the experiment. Even though pupillary response has shown to be sensitive for changes in MWL, this reactivity to changing states of workload and arousal is only one of three underlying mechanisms that induce changes of pupil size [
48]. The remaining two are the pupil light reflex and the pupil near or accommodation reflex, of which both could have an influence on this study. More research has been done on the pupil light reflex focusing on changes of pupil size due to changing light conditions, resulting in a pupillary constriction for increases in brightness and vice versa, a dilation for increasing darkness [
67]. During the experiment, we focused on keeping lighting conditions constant. Therefore, three sources of light were applied on the assembly station, two directly above each working area (see assembly object in
Figure 3) and one central over the assembly station at the ceiling. Thus, the room should have been evenly illuminated. Even with promising approaches to eliminate the influence of light on pupillary response during mental workload, further research and additional light sensors are needed to use them [
68]. While lighting and brightness should not be an issue for the used assistance systems (tablet was set to a medium brightness for good visibility, but no disturbing brightness), the pupil near reflex could have an influence on the comparison due to decreased distance for the information available on the AR-glasses. The pupil near reflex “is certainly the least studied, and perhaps the least understood of all pupil responses” [
48] (p. 10). It describes the dilation of the pupil focusing on distant objects and the constriction focusing on near ones. During the experiment, most visual cues were presented in a range of 1.5 m. For AR-glasses, an optical see-through version was chosen with a small projection area in the top right corner that did not interfere too much with the user’s natural field of view [
69]. Thus users of the AR-glasses could have an overall smaller pupil size due to the shorter fixation distance for the presented instructions. Noticeable differences only occurred during M1 with 3.29 mm against 3.43 mm (Tablet) and 3.54 mm (Paper) (
Table 4). Still, the participants in this group showed the highest amount of overload during M1 (34.53%,
Table 5), therefore the pupil near reflex seems to be a negotiable issue for this study.
Including EEG and neuroimaging techniques might be a way to obtain a closer look at the brain at work, but will also be more intrusive and lead to additional questions like how to handle latency between different techniques, which signals refer to task specific changes and which ones to wandering thoughts, and how to handle the comparatively low number of task repetitions in a more field-like environment.