Tài liệu hạn chế xem trước, để xem đầy đủ mời bạn chọn Tải xuống
1
/ 283 trang
THÔNG TIN TÀI LIỆU
Thông tin cơ bản
Định dạng
Số trang
283
Dung lượng
4,57 MB
Nội dung
Multimodal Data Fusion for Cyber-Physical-Human Systems A thesis submitted in fulfilment of the requirements for the degree of Master of Engineering Lars Joyce Planke BEng (Electrical) (Hon 1st Class) (RMIT University) School of Engineering College of Science, Technology, Engineering and Maths RMIT University May 2021 Declaration I certify that except where due acknowledgement has been made, this work is that of the author alone; this work has not been submitted previously, in whole or in part, to qualify for any other academic award; the content of the thesis is the result of work, which has been carried out since the official commencement date of the approved research program; any editorial work, paid or unpaid, carried out by a third party is acknowledged; and, ethics procedures and guidelines have been followed. I acknowledge the support I have received for my research through the provision of an Australian Government Research Training Program Scholarship. Lars Joyce Planke 20 May 2021 I Acknowledgements In conducting my Master of Engineering research program, I would firstly like to express my gratitude to my supervisors Prof. Roberto (Rob) Sabatini and Dr. Alessandro (Alex) Gardi, in which their advice, guidance and aspirations have been of utmost importance in completing a successful research project. I would like to give my thanks to Rob for identifying a potential in me, opening up invaluable opportunities and aspiring me to strive for excellence. I would also like to thank Alex for providing time and thoughtful advice throughout the project. A special thanks goes to my friends and colleagues at RMIT University, including Yixiang Lim, Nichakorn Pongsarkornsathien, Sam Hilton, Rohan Kapoor, Suraj Bijjahalli and Federico Rivalta. Their accompanying presence have made this endeavour highly rewarding, with stimulating interactions that have sparked valuable insights and helped progress my research. I would also like to give many thanks to Dr. Neta Ezer from Northrop Grumman Corporation for providing helpful feedback and supporting this research project. Also, thanks to Trevor Kistan from Thales Australia for helping me conduct this research and providing feedback. I would also like to acknowledge and express my gratitude to RMIT University for selecting me for a RMIT Research Stipend Scholarship. This research project has been conducted amid a historic pandemic, and a particular gratitude goes to my partner Bella Reid and her family for providing extra support through unusual and uncertain times, without which this research would not have been as successful. Finally, I would to like to give my heartfelt appreciation to all my friends and family in Norway and Australia. No matter the physical distance I can always rely on their encouragement, care and constant source of support. II Table of Contents Declaration Acknowledgements I II List of Figures VIII List of Tables XI XII 1 List of Acronyms and Abbreviations Executive Summary Background 3 Research gaps 6 1.2.1. Scope _ 10 Research aim and question _ 11 Research objectives _ 12 Overview of research methodology 13 1.5.1. General project methodology 13 1.5.2. Experimental activities _ 14 Thesis outline 16 References 17 Introduction _ 23 Human centred systems _ 24 Architecture for future aerospace systems 28 2.3.1. HMI2 in the CNS+A context 28 2.3.2. Adaptive systems and CHMS _ 30 2.3.3. CHMS in future aerospace applications 32 2.3.4. Evolutionary paths for future multidomain traffic management _ 37 Cognitive modelling and measurements 38 III 2.4.1. Conceptual relationship between task load, performance and MWL _ 38 2.4.2. Cognitive modelling and definitions _ 41 2.4.3. Types of methods for measuring MWL and generating task loads _ 43 Sensors and methods for real‐time measurement of MWL 49 2.5.1. Electroencephalogram (EEG) 49 2.5.2. Eye activity tracking 62 2.5.3. Cardiac activity _ 64 2.5.4. Control input measurements 66 2.5.5. Other sensor measurements of MWL 67 2.5.6. Multimodality fusion _ 68 2.5.7. Generalised inference models _ 78 Conclusions _ 80 References 82 Introduction _ 95 Design considerations for sensing and estimation _ 96 3.2.1. Top‐level requirements for sensing and estimation _ 97 3.2.2. Processes for sensing and estimation 98 Research methodology _ 101 3.3.1. General project methodology _ 102 3.3.2. Task load generation 104 3.3.3. Participant inclusion criteria and ethics _ 109 3.3.4. Sensor equipment 110 3.3.5. Software infrastructure 115 3.3.6. Data analysis 117 Experimental approach and design 119 3.4.1. Experimental Activity 1 119 3.4.2. Experimental Activity 2 122 3.4.3. Experimental Activity 3 126 Conclusion _ 128 IV References _ 129 Introduction 131 OTM UAS test case 132 4.2.1. Participants _ 132 4.2.2. Mission concept 133 4.2.3. Experimental procedure _ 134 Measurement methods _ 135 4.3.1. Eye tracking data processing 135 4.3.2. EEG data processing 136 4.3.3. Controller input processing _ 137 4.3.4. Secondary task performance index _ 137 Data analysis _ 138 4.4.1. ANOVA test _ 138 4.4.2. Correlations between features 139 Results 139 4.5.1. ANOVA test _ 139 4.5.2. Correlation between features _ 144 Discussion _ 147 Conclusion _ 151 References _ 152 Introduction 153 Experimental procedure and setup 154 Participants 155 MATB scenario design 156 V Data collection and processing _ 157 5.5.1. Subjective measure _ 158 5.5.2. MATB performance measures 158 5.5.3. Physiological and behavioural measurements 158 5.5.4. Final inference model _ 164 5.5.5. Network during experiment 168 Data analysis _ 170 5.6.1. Analysing the individual features using the CC 170 5.6.2. Analysing the inference models _ 171 Results 171 5.7.1. Calibrating and analysing the EEG model 172 5.7.2. Individual results from all features _ 175 5.7.3. Correlations between all features 180 5.7.4. Results from calibrating and validating the ANFIS _ 181 Discussion _ 190 5.8.1. Discussion of results 191 5.8.2. Discussion of contribution 198 Conclusion _ 207 References 209 Introduction 213 Experiment procedure 214 Participants 214 MATB scenario design 215 Data collection and processing _ 216 6.5.1. Networking during the experiment 217 Data analysis _ 219 6.6.1. Round 1 analysis _ 219 6.6.2. Round 2 analysis _ 220 VI 6.6.3. Threshold criterion _ 220 Results 221 6.7.1. Round 1 results 221 6.7.2. Round 2 results 223 Discussion _ 232 6.8.1. Discussion of methodology design and networking considerations _ 232 6.8.2. Discussion of results 234 6.8.3. Discussion of contribution 239 6.8.4. Discussion of multimodal inference of MWL for CHMS _ 241 Conclusion _ 244 References 246 Introduction 249 Synthesis of experimental findings 249 Conclusion _ 254 7.3.1. Objectives achieved _ 256 Recommendations for future research _ 259 Appendix A – Subject Specific Feature Combination 261 Appendix B – Neuropype Pipeline Designer 264 Appendix C – Results from Normality Test 266 Appendix D – List of Publications 268 VII List of Figures Figure 1.1 Fundamental concept of the Cognitive Human Machine System (CHMS)…………………5 Figure 1.2 The general project methodology………………………………………………………………….……… 14 Figure 1.3 General methodology for Experimental Activity 1…………………………………………….…….15 Figure 2.1 Sheridan scale……………………………………………………………………………………………….……… 26 Figure 2.2 Full CHMS framework ……………………………………………………………………………….……… 31 Figure 2.3 VPA system architecture………………………………………………….…………………………………….33 Figure 2.4 Integrated air‐ground Concepts of Operation for SPO and UAS remote control .35 Figure 2.5 Layers of control for UAS and SPO…………………………………………………….……………………36 Figure 2.6 Third generation flight deck concept by Thales……………………………………………….……. 37 Figure 2.7 Evolutionary paths for a multidomain air and space transport operation.….… 38 Figure 2.8 Inverted U model……………………………………………………………………………………… …………39 Figure 2.9 Relationship between task load, performance and MWL……………………………………….40 Figure 2.10 Conceptual relationship between MWL and SA………………………………………………………41 Figure 2.11 Tasks for the MATB program………………………………………………………………………………….44 Figure 2.12 International 10‐20 system for electrode placement.………………………………………… …50 Figure 2.13 (a) Referential montage; (b) Differential amplifier circuit………………………….……………51 Figure 2.14 A simulated head for finding oscillating voltage sources…………………………….… …… 53 Figure 2.15 Signal processing method for a straightforward oscillatory BCI………………………………56 Figure 2.16 EEG signals spatially filtered using CSP algorithms…………………………………………………58 Figure 2.17 Filter Bank Common Spatial Pattern (FBCSP)………………………………………………….………59 Figure 2.18 A generic signal processing for ERP based BCI……………………………………………….……….61 Figure 2.19 QRS complex…………………………………………………………………………………………………… 65 Figure 2.20 Example of different modalities that can be used for cognitive load………………… 70 Figure 2.21 High level data fusion for multimodal MWL estimation………………………………… .71 Figure 2.22 Example of an architecture of a NFS………………………………………………………………………76 Figure 3.1 Basic configuration of CHMS……………………………………………………………………….…………96 Figure 3.2 Fundamental components of sensing and estimation………………………………… ………98 Figure 3.3 The general project methodology……………………………………………………………… …….103 Figure 3.4 OTM UAV wildfire detection scenario……………………………………………………… … …….106 Figure 3.5 Tasks for the MATB program………………………………………………………………………… … 107 Figure 3.6 (a) actiCAP Xpress form BrainProducts; (b) V‐Amp amplifier with USB cable… ……111 Figure 3.7 GP3 eye tracker…………………………………………………………………………………………… …… 112 Figure 3.8 Zephyr Bioharness 3………………………………………………………………………………….… …….112 VIII Figure 3.9 Logitech Extreme 3D Pro………………………………………………………………….………………….113 Figure 3.10 Example of eye tracker mistakenly detecting features of the EEG electrodes as… 114 Figure 3.11 (a) Cloth that is placed to cover the electrodes of the EEG; (b) Cloth preventing… 114 Figure 3.12 Basic data flow for physiological measurement collection……………………………………115 Figure 3.13 General methodology for activity 1………………………………………………………………………119 Figure 3.14 Overall methodology for Experimental Activity 1……….……………………………………… 121 Figure 3.15 Multimodal data fusion with testing of multiple inference models……………………….123 Figure 3.16 Detailed methodology for offline development and analysis of inference………… 124 Figure 3.17 Offline calibration and validation of ANFIS model……………………………………………… 125 Figure 3.18 Overall methodology for Experimental Activity 3…………………………………………… 127 Figure 4.1 Mission concept illustrating the Unmanned Aerial Vehicles (UAVs) bounding…… 134 Figure 4.2 Subjective ratings: (a) subjective workload of all participants, grouped by the…….141 Figure 4.3 (a) Mean task index of all participants (normalized), grouped by the scenario…… 142 Figure 4.4 Physiological measures: (a) mean scan pattern entropy of all participants………….143 Figure 4.5 Mean EEG index of all participants…………………………………………………………………….…144 Figure 4.6 Comparison for one participant between the task index, SPE and EEG index……… 145 Figure 5.1 (a) Experimental setup while participant performs MATB tasks. A: EEG cap……… 155 Figure 5.2 MATB task load profile for the experiment……………………………………………………………156 Figure 5.3 Diagram showing all the categories and subcategories of data collection…………….157 Figure 5.4 Feature extraction using FBSPoC and continuous classification using ridge………….159 Figure 5.5 Pipeline for offline calibration and validation of EEG model…………………………… 160 Figure 5.6 Four features extracted from the eye tracking data…………………………………………… 162 Figure 5.7 ROI for calculating simple (a) and complex; (b) scan pattern entropy……………….… 163 Figure 5.8 Network diagram for processing and collection of data……………………………………….169 Figure 5.9 Protocol for synchronising measurements……………………………………………………………170 Figure 5.10 Result for participant 11 after calibrating the EEG model…………………………………… 172 Figure 5.11 Results from the EEG pipeline for participant six………………………………………………… 174 Figure 5.12 (a) EEG measure for participant 4; (b) scan pattern entropy for participant 2………176 Figure 5.13 (a) Blinks per minute for participant 16; (b) pupil diameter for participant 2……… 176 Figure 5.14 (a) Dwell time for participant 5; (b) Heart rate for participant 3……………………………177 Figure 5.15 Control inputs for participant 14………………………………………………………………………….177 Figure 5.16 Calibration and validation on separate halves of the data set……………………………….182 Figure 5.17 Results after calibrating and validating the ANFIS model for participant 4……….… 189 Figure 6.1 Overall methodology……………………………………………………………………………………… 214 Figure 6.2 Task profile for Round 1 of the MATB scenario…………………………………………… …… 215 Figure 6.3 Task profile for the online validation round in Round 2…………………………………………216 IX Chapter 7. Synthesis, Conclusion and Recommendations 254 implementation of non‐contributing features was not detrimental to the performance of the ANFIS model when using it for multimodal fusion, as the use of multiple features can prove vital in the reliability of the inference of MWL. Conclusion With increasingly higher levels of automation in aerospace CPS it is imperative that human operators maintain the required level of SA in order to contribute effectively to both strategic and tactical decision‐making processes. To ensure that the performance of the system and the MWL of the human operator are maintained at an acceptable level, one possible approach is to introduce real‐time adaptation in HMI2. Current aerospace systems and interfaces are limited in adaptability where the authority for reconfiguration and task allocation in current HMI2 is manually controlled by the human operator. Henceforth, a CHMS addresses these issues with a cyber‐physical human design that provides dynamic real‐time system adaptation. Nevertheless, to reliably drive adaptation of aerospace decision support systems there is a need to accurately and repeatably estimate cognitive states, in particular MWL, in real‐time. Henceforth this research has studied methods for sensing physiological and behavioural responses associated with MWL and used the corresponding features to provide a multimodal inference of MWL. Previous work related to the measurement of MWL has demonstrated the capability of supervised ML models to improve the estimation of MWL. Nevertheless, classification models using multiple modalities have generally been implemented with discrete classification (i.e. resting, low and high), and where the analysis has been conducted using offline calibration and offline validation. The use of ML models that employ a continuous regression approach for the multimodal inference of MWL is less implemented. While some previous work has included the use of a NFS to fuse the data from multiple modalities, the accurate online performance of an ANFIS model for the multimodal inference of MWL has not previously been reported. This includes the thorough investigation of the performance of the model from an online validation, with an analysis of the performance of the model based on using different feature combinations. Multimodal Data Fusion for Cyber‐Physical‐Human Systems 255 The work conducted as part of this research included an exploratory experimental activity (Experimental Activity 1) that implemented and analysed an EEG index during a OTM UAV wildfire detection scenario. The index included previously proven features of MWL, including the changes in the theta and alpha band, and showed to be sensitive to changes in MWL in the complex task scenario. Moreover, a straightforward data fusion approach showed that fusing the EEG index and scan pattern entropy features gave the highest correlation with the secondary task performance measure (CC = 0.73 ± 0.14). The following more comprehensive Experimental Activities 2 and 3 conducted as part of this research consisted of the development and offline/ online validation of a multimodal inference model of MWL that was experimentally tested during a MATB scenario. While the terms offline validation and online validation were adopted in this thesis, the terms were used to reflect the language used in the literature and to convey that the models were tested with “unseen” data in offline and online processing. Nonetheless, these experimental activities were exploratory studies into the development of a multimodal inference model of MWL. These experimental activities involved a rigorous analysis of subject specific inference models and included using a subject specific EEG model and an ANFIS model for fusing the multimodal physiological and behavioural features. The multimodal inference model implemented in Experimental activity 2 and 3 included an ANFIS that used FCM as an initial clustering method. Once the initial clustering was generated, a further tuning process was performed that used backpropagation to tune the input membership function and least squares estimation to tune the output membership functions (see Section 5.5.4.1 for further detail). The results from the offline calibration and validation (Experimental Activity 2), of the several multimodal inference models of MWL, demonstrated that the average from the best performing subject specific feature combinations gave the lowest error with the task level with a MAE = 0.28. Nonetheless, the results from using all the seven features in the combination showed a MAE = 0.36. While the former demonstrated a smaller error, the latter showed quite a comparable result, hence demonstrating that the use of the other non‐contributing features was not as detrimental to the ANFIS when performing multimodal inference of MWL. The final Experimental Activity 3 included the online validation of 11 selected ANFIS models as determined in the previous activity. In addition to this the online validation of ANFIS Chapter 7. Synthesis, Conclusion and Recommendations 256 models 1‐5 (containing different feature combinations of eye activity and control input features) where tested for cross‐session inference of MWL on half of the participants in Round 1, while ANFIS models 10‐11 used a subject specific feature combination and were tested on half of the participants in Round 2. As in the previous case the features were thoroughly assessed using the pairwise correlation with the task level. The average showed in both Experimental Activities 2 and 3 to be consistent with the results in previous studies. The results from the online cross‐session validation of ANFIS models 1‐5 in Round 1 showed that all models performed well, with ANFIS model 2 showing the lowest error with a MAE = 0.63. The ANFIS models 1‐5 performed equally well across all subjects in Round 2 with ANFIS model 3 showing a MAE = 0.67 and CC = 0.71. The ANFIS models 6‐11 (performed in Round 2) included the measure from the EEG model that showed a discrepancy with an offset. This equally resulted in an offset in the final online inference of the ANFIS models thus resulting in a large error. Nonetheless, the efficacy of the EEG model and resulting ANFIS models could be seen with the normalised pairwise correlation between the outputs and the task level. For ANFIS models 6‐9, the results showed a good pairwise correlation in the range of 0.61 to 0.69, while ANFIS model 11 that implemented the subject specific feature combination showed a CC = 0.77, which was the best result among all the models tested. Henceforth these results demonstrated the ability for multimodal data fusion from features extracted from EEG, eye activity and control inputs to produce an accurate and repeatable inference of MWL. In particular this included ANFIS model 1‐5, that showed to consistently produce a repeatable inference of MWL during within‐ and cross‐session validation. The investigation of multimodal fusion for MWL inference has assisted in corroborating the viability of real‐time system adaptation in future aerospace CPHS architectures. 7.3.1 Objectives achieved The aim of this research was the development of an accurate and repeatable inference model for MWL estimation as needed for a CHMS. In working towards this the following objectives were achieved. OBJ A comprehensive literature review on MWL measurements, EEG, ECG, eye activity tracking, control inputs, human factors, adaptive automation, closed Multimodal Data Fusion for Cyber‐Physical‐Human Systems 257 loop system adaptation, passive BCI, HMI2, CNS+A, multimodal data fusion, machine learning and ANFIS: The real‐time inference of MWL is a multidisciplinary field that requires the review of many different areas. As such this research has completed a detailed review of multiple areas ranging from the theoretical relation between various cognitive states, to the technology of the sensors performing the physiological and behavioural measurements and the analysis of this data for producing a final multimodal inference of MWL. In addition to this a review was conducted on the CHMS and the application of it for future aerospace operations. OBJ Identify methods for multimodal data fusion and corresponding data labelling/ target value strategies: As part of the review this objective was more specifically emphasised. This objective was achieved with a detailed overview of the various multimodal data fusion techniques and an investigation of applicable labelling techniques as used for the inference model. As supervised ML techniques were implemented a labelling strategy was identified and implemented. OBJ Determine and experimentally verify physiological and behavioural measurements associated with MWL: Extensive reviews have identified that measures of MWL are not universally valid for all task scenarios. Henceforth, this research has thoroughly analysed a number of different physiological and behavioural features associated with MWL. The results from the experimental activities has verified several strong correlations between the respective measurements of MWL and other objective measures of MWL (i.e. task performance or task analysis). Most of the experimentally tested measures of MWL have shown to be consistent with previously reported findings in the literature. OBJ Select and experimentally implement a subject specific machine learning technique most suited for continuously inferring MWL at frequent intervals based on multimodal data: Chapter 7. Synthesis, Conclusion and Recommendations 258 To achieve a repeatable and accurate inference of MWL, two ML models were selected and later implemented for experimental testing. This included the use of a supervised ML models for the EEG data as well as an ANFIS for fusing the multimodal data. The results from the ANFIS model demonstrated an accurate inference of MWL at regular 1 second intervals, and with the EEG model demonstrating an inference of MWL at 3 second intervals. OBJ Perform a thorough analysis from offline calibration and validation of a multimodal inference model of MWL and analyse performance from using different generic and subject specific feature combinations: As completed in Experimental Activity 2, a thorough offline calibration and validation was conducted using a multimodal ANFIS model that was calibrated using a number of feature combinations, including a generic and a subject specific combination. OBJ Perform online validation of the multimodal inference models of interest, including online cross-session validation: As completed in Experimental Activity 3, an online validation of the multimodal inference model of MWL was performed and showed for some models to produce accurate and repeatable inference of MWL based on real‐time data. An online cross‐session validation of five of the multimodal models demonstrated the ability for cross‐session inference of MWL. OBJ Analyse elements of explainability of the inference of MWL from the ML model implemented: As part of the pairwise correlation analysis with the respective features and the task level, the multimodal ANFIS model performed well in line with the preceding correlation analysis. Moreover, an analysis conducted after the online validation of the ANFIS models demonstrated which features correlated strongest with the real‐ time output, thus providing some level of explainability in terms of the features the model emphasised in the final multimodal inference of MWL. Multimodal Data Fusion for Cyber‐Physical‐Human Systems 259 OBJ Create and implement a protocol for rapid calibration of a multimodal inference model: For an operational CHMS, calibration of the various models will need to be performed efficiently. As part of Experimental Activity 3, a rapid calibration protocol was developed where the subject specific inference models could be calibrated and then validated in real‐time during a single session. Recommendations for future research The following key points have been identified as recommendations for further research activities in the area of real‐time multimodal MWL estimation: A major challenge within this research area is achieving reliable cross‐task classification/ inference from a calibrated multimodal model. Hence future research should further investigate the capability of calibrating a multimodal model on one task and validating the multimodal model on another task. As part of future research, it would be viable to take the multimodal model developed in Chapter 6 and assessing its ability to infer MWL on another task; Building on the previous point, the ability for cross‐task inference is dependent on the measurement to be task independent. Thus, future research should focus on studying physiological and behavioural measure of MWL that are task independent. This includes exploring new/ other task independent physiological and behavioural measures (i.e. facial expression) for their sensitivity to changes in MWL, and their effect on performing multimodal fusion; Having accurate and frequent labelled data is crucial when implementing supervised ML techniques, however knowing the “ground truth” of the humans perceived MWL is one of the major challenges. Henceforth, an area for further research would be implementing a different objective measure of MWL, such as task‐based measures or reserving other well‐established physiological/ behavioural measures as data labels for calibration and validation. This additionally includes implementing a more objective assessment of the model’s performance by using one measure for labelling the data and another objective measure for validating the model’s performance; Chapter 7. Synthesis, Conclusion and Recommendations 260 Many of the ML models techniques lack in the explainability of the final estimation of MWL. Future research should thus explore the explainability of inference models, such that the operator’s cognitive state can be explained at important junctures in real‐time. This includes studying how an operator can be prompted regarding a certain level of inferred MWL whilst conducting a certain task; While the experimental activities implemented some limited artifact rejection, future research should considerably extend the reliability and maturity of real‐time artifact removal from the various sensors; While this research focused on the multimodal sensing and estimation of MWL, future research should further study the implementation of this for driving real‐time system adaptation in conventional and emerging aerospace concepts such as ATC, SPO or UTM. This should include studying how the dynamic system adaptation affects the operation of the system and should investigate if the trust of the operation is maintained. Appendix A Subject Specific Feature Combination Table A.1 Subject specific feature combination with the Controller Input (CI) feature (see Section 5.5.3.4) (not including features with pairwise correlation less than CC=0.45) Subject Best features Best features Best features Best features Best features SPE, CI EEG, CI, SPE EEG, CI, SPE, DIA EEG, CI, SPE, DIA, Dwell N/A DIA, EEG DIA, EEG, SPE DIA, EEG, SPE, CI DIA, EEG, SPE, CI, HR N/A CI, SPE CI, SPE, BPM CI, EEG, SPE, BPM CI, EEG, SPE, BPM, HR SPE, CI EEG, SPE, CI EEG, SPE, CI, Dwell EEG, SPE, CI, Dwell, Dia CI, EEG, SPE, BPM, HR, DIA N/A SPE, CI SPE, CI, Dia SPE, CI, DIA, EEG SPE, EEG, CI, DIA, Dwell EEG, CI EEG, CI, SPE EEG, CI, SPE, DIA EEG, CI, SPE, DIA, HR EEG, SPE, DIA, Dwell, BPM N/A EEG, CI EEG, CI, SPE EEG, CI, SPE, BPM N/A N/A SPE, DIA SPE, DIA, CI SPE, DIA, CI, EEG SPE, DIA, CI, EEG, Dwell CI, SPE CI, SPE, EEG CI, SPE, EEG, Dwell CI, SPE, EEG, Dwell, DIA SPE, DIA, click, EEG, Dwell, BPM N/A 10 EEG, CI EEG, CI, SPE EEG, CI, SPE, DIA EEG, CI, SPE, DIA, BPM 11 EEG, CI EEG, CI, DIA EEG, CI, DIA, BPM EEG, CI, DIA, BPM, Dwell 12 CI, SPE CI, SPE, Dia CI, SPE, DIA, EEG CI, EEG, SPE, DIA, BPM 13 SPE, CI SPE, CI, EEG SPE, CI, EEG, Dwell N/A EEG, CI, SPE, DIA, BPM, Dwell EEG, CI, DIA, BPM, Dwell, SPE CI, EEG, SPE, DIA, BPM, HR N/A 14 EEG, CI EEG, CI, SPE EEG, CI, SPE, Dwell EEG, CI, SPE, Dwell, DIA N/A 15 SPE, CI EEG, CI, SPE EEG, CI, SPE, DIA EEG, CI, SPE, DIA, BPM 16 CI, SPE CI, SPE, BPM CI, SPE, BPM, EEG EEG, CI, SPE, BPM, DIA 17 CI, SPE CI, SPE, DIA CI, SPE, DIA, EEG EEG, CI, SPE, DIA, BPM EEG, CI, SPE, DIA, BPM, Dwell EEG, CI, SPE, BPM, DIA, Dwell EEG, CI, SPE, DIA, BPM, Dwell 261 262 Table A.2 Subject specific feature combinations without the Controller Input (CI) feature (see Section 5.5.3.4) (not including features with pairwise correlation less than CC=0.45) Participant Best features Best features Best features Best features All features EEG, SPE EEG, SPE, DIA EEG, SPE, DIA, Dwell N/A N/A EEG, DIA EEG, DIA, SPE EEG, DIA, SPE, HR EEG, DIA, SPE, HR, dwell N/A SPE, BPM EEG, SPE, BPM EEG, SPE, BPM, HR EEG, SPE, BPM, HR, DIA EEG, SPE EEG, SPE, Dwell EEG, SPE, Dwell, DIA N/A EEG, SPE, BPM, HR, DIA, Dwell N/A SPE, DIA EEG, SPE, DIA EEG, SPE, DIA, Dwell EEG, SPE, DIA, Dwell, BPM N/A EEG, SPE EEG, SPE, DIA EEG, SPE, DIA, HR N/A N/A EEG, SPE EEG, SPE, BPM N/A N/A N/A SPE, DIA SPE, DIA, EEG SPE, DIA, EEG, Dwell SPE, DIA, EEG, Dwell, BPM N/A EEG, SPE EEG, SPE, Dwell EEG, SPE, Dwell, DIA N/A N/A 10 EEG, SPE EEG, SPE, DIA EEG, SPE, DIA, BPM EEG, SPE, DIA, BPM, Dwell N/A 11 EEG, DIA EEG, DIA, BPM EEG, DIA, BPM, Dwell EEG, DIA, BPM, Dwell, SPE N/A 12 SPE, DIA EEG, SPE, DIA EEG, SPE, DIA, BPM EEG, SPE, DIA, BPM, HR N/A 13 EEG, SPE EEG, SPE, Dwell N/A N/A N/A 14 EEG, SPE EEG, SPE, Dwell EEG, SPE, Dwell, DIA N/A N/A 15 EEG, SPE EEG, SPE, DIA EEG, SPE, DIA, BPM EEG, SPE, DIA, BPM, Dwell N/A 16 SPE, BPM EEG, SPE, BPM EEG, SPE, BPM, DIA EEG, SPE, BPM, DIA, Dwell N/A 17 SPE, Dias EEG, SPE, DIA EEG, SPE, DIA, BPM EEG, SPE, DIA, BPM, Dwell N/A 263 Table A.3 Selection of best performing subject specific feature combinations (comb.) for ANFIS models 10 and 11 based on results from offline validation. (CI indicates the Controller Input feature (see Section 5.5.3.4)) Subject 10 11 12 13 14 15 16 17 Custom ANFIS 10 Best model - no CI Best 3 comb. Best 3 comb. Best 4 comb. Best 2 comb. Best 4 comb. All 6 comb. Best 2 comb. Best 4 comb. All 6 comb. Best 4 comb. Best 5 comb. All 6 comb. Best 2 comb. Best 4 comb. Best 2 comb. Best 2 comb. Best 3 comb. Custom ANFIS 11 Best model - with CI Best 3 comb. Best 3 comb. Best 5 comb. All 7 comb. Best 3 comb. Best 2 comb. Best 3 comb. Best 4 comb. Best 2 comb. Best 5 comb. Best 2 comb. Best 6 comb. Best 2 comb. Best 3 comb. Best 2 comb. Best 3 comb. Best 2 comb. Appendix B Neuropype Pipeline Designer Figure B.1 EEG pipeline implemented in Pipeline Designer for offline calibration and validation for Experimental Activity 2. 264 265 Figure B.2 EEG pipeline implemented in Pipeline Designer for offline calibration and online validation during Round 2 in Experimental Activity 3. Appendix C Results from Normality Test The following are the results from testing for normality for the outputs from the ANFIS models for both Experimental Activity 2 (Chapter 5) and Experimental Activity 3 (Chapter 6). As evident in the respective tables below all results passed the normality test. Table C.1 Testing for normality for all the outputs from the ANFIS models 1‐11 (validated on first half of data (Section 5.7.4.2)) and respective subjects in Session 1. All outputs from the multimodal ANFIS models passed the normality test. (E represents times ten raised to the power). 266 267 Tabel C.2 Testing for normality for all the outputs from the ANFIS models 1‐5 and respective subjects in Session 2 Round 1 (Section 6.7.1.1). All outputs from the multimodal ANFIS model passed the normality test. (E represents times ten raised to the power). ANFIS 4.7 E‐47 1.4 E‐77 1.5 E‐33 7.6 E‐35 5.9 E‐68 ANFIS ANFIS ANFIS ANFIS 6.7 E‐107 1.6 E‐123 3.4 E‐130 7.8 E‐139 4.8 E‐89 1.5 E‐97 4.8 E‐103 1.6 E‐44 1.7 E‐95 3.9 E‐88 5.8 E‐99 1.6 E‐46 3.7 E‐27 6.1 E‐96 1.3 E‐21 2.1 E‐123 1.5 E‐91 4.0 E‐56 1.8 E‐51 2.3 E‐65 7.1 E‐109 2.5 E‐104 2.4 E‐251 3.1 E‐160 4.4 E‐217 Table C.3 Testing for normality for all the outputs from the ANFIS models 1‐11 and respective subjects in Session 2 Round 2 (Section 6.7.2.1). All outputs from the multimodal ANFIS model passed the normality test. (E represents times ten raised to the power). ANFIS ANFIS ANFIS ANFIS ANFIS ANFIS ANFIS ANFIS ANFIS 7.0 E‐09 8.8 E‐14 7.5 E‐36 3.1 E‐40 1.1 E‐33 2.4 E‐12 2.0 E‐58 1.4 E‐23 2.0 E‐18 ANFIS 10 n/a ANFIS 11 n/a 4.3 E‐32 1.5 E‐12 1.5 E‐51 5.6 E‐29 2.3 E‐44 0 0 1.6 E‐27 0 0 0 2.0 E‐28 3.4 E‐39 1.7 E‐09 5.5 E‐26 7.1 E‐64 4.0 E‐16 8.1 E‐39 9.5 E‐41 8.4 E‐43 6.6 E‐47 1.4 E‐47 6.8 E‐60 3.1 E‐44 2.7 E‐55 6.0 E‐41 1.9 0 0 E‐106 4.6 2.0 0 E‐227 E‐47 5.6 7.6 0 E‐145 E‐35 3.9 8.8 0 E‐226 E‐111 5.5 9.9 n/a E‐175 E‐324 8.4 0 n/a E‐29 4.5 E‐77 2.7 E‐78 3.9 E‐37 5.5 E‐07 6.0 E‐21 2.8 E‐66 0 0 9.9 E‐83 0 1.4 E‐122 3.2 E‐57 3.6 E‐45 7.0 E‐45 1.7 E‐12 8.0 E‐58 4.1 E‐52 5.1 E‐41 2.4 E‐23 1.5 E‐49 1.4 E‐47 4.3 E‐49 2.0 E‐23 2.5 E‐11 7.7 E‐16 4.5 E‐20 3.6 E‐67 3.0 E‐58 6.1 E‐86 3.0 E‐39 1.6 E‐71 1.9 E‐42 8.1 E‐26 3.6 E‐23 1.9 E‐83 2.3 E‐26 10 1.6 E‐19 7.2 E‐45 6.6 E‐25 1.3 E‐04 2.0 E‐70 11 5.9 E‐48 2.7 E‐33 3.3 E‐35 1.5 E‐27 1.0 E‐65 6.6 E‐92 1.5 E‐120 3.6 E‐40 12 3.5 E‐18 1.8 E‐35 7.7 E‐21 3.7 E‐43 8.7 E‐91 1.2 E‐34 2.4 E‐21 2.2 E‐98 4.7 E‐157 0 0 0 2.0 E‐50 0 0 0 0 0 n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a Appendix D List of Publications As part of this research project, the author has produced the following publications. Lead author publications: L J Planke, A. Gardi, R. Sabatini, T. Kistan and N. Ezer, "Online Multimodal Inference of Mental Workload for Cognitive Human Machine Systems," Computers, vol. 10, no. 6, p. 81, June 2021 L J Planke, Y. Lim, A. Gardi, R. Sabatini, T. Kistan, and N. Ezer, "A Cyber‐Physical‐ Human System for One‐to‐Many UAS Operations: Cognitive Load Analysis," Sensors, vol. 20, no. 19, p. 5467, September 2020 Co‐author publications: N. Pongsakornsathien, Y. Lim, A. Gardi, S. Hilton, L J Planke, R. Sabatini, T. Kistan, N. Ezer, "Sensor Networks for Aerospace Human‐Machine Systems," Sensors, vol. 19, no. 16, p. 3465, August 2019 268 ... further developing an EEG index and implementing a? ?data? ?fusion? ?approach with the EEG Multimodal? ?Data? ?Fusion? ?for? ?Cyber? ? ?Physical? ? ?Human? ?Systems? ? 15 index and an eye activity measure. The EEG index and? ?data? ?fusion? ?approach were analysed ... However, it was concluded that the physiological and performance measures improved the system adaptation over using only system error? ?for? ?driving the adaptation. Multimodal? ?Data? ?Fusion? ?for? ?Cyber? ? ?Physical? ? ?Human? ?Systems? ? 9 In terms? ?of? ?estimating MWL, a continuous regression measure? ?of? ?MWL is arguably more ... scale, should be specified at the various phases? ?of? ?the flight profile. Multimodal? ?Data? ?Fusion? ?for? ?Cyber? ? ?Physical? ? ?Human? ?Systems? ? 35 Figure 2.4. Integrated Air‐Ground Concepts? ?of? ?Operation? ?for? ?SPO and UAS remote control [29].