UNCORRECTED PROOF Applied Ergonomics xxx (xxxx) xxx-xxx Contents lists available at ScienceDirect Applied Ergonomics journal homepage: http://ees.elsevier.com Detecting cognitive hacking in visual inspection with physiological measurements Wenyan Huang, Xiaoyu Chen, Ran Jin, Nathan Lau ∗ The Grado Department of Industrial and Systems Engineering, Virginia Tech, Blacksburg, USA ARTICLE INFO Keywords Additive manufacturing Cognitive hacking Cybersecurity EEG Eye-tracking Inspection ABSTRACT Cyber threats are targeting vulnerabilities of human workers performing tasks in manufacturing processes, in- cluding visual inspection to bias their decision-making, thereby sabotaging product quality. This article examines the use of priming as a form of “cognitive hacking” to adversely affect quality inspection decisions in manufac- turing, and investigates physiological measurements as means to detect such intrusion. In a within-subject design experiment, twenty participants inspected surface roughness of a manufactured component with and without ex- posure to priming on the display of an inspection logging system. The results show that the presence of primes impacted accuracy on surface roughness, cortical activities at parietal lobe P4, and eye gaze for inspecting com- ponents. The experiment provides supporting evidence that basic hacking of a worker display can be an effective method to alter decision making in inspection. The fndings also illustrate that cortical activities and eye gaze can be useful indicators of cognitive hacking. A major implication of the study results is that physiological indicators can be effective at revealing unconscious cognitive infuence in visual inspection. 1. Introduction Visual inspection plays an important role in safety and productiv- ity for a wide range of industries. An illustrative safety case is the 1989 crash of the aircraft DC-10-10 killing 111 passengers that the Na- tional Transportation Safety Board (1990, 102) determined “the proba- ble cause of this accident was the inadequate consideration given to hu- man factors limitations in the inspection and quality control procedures …“. Latorella and Prabhu (2000) presented other aircraft accidents attributable to inspection failures. With regards to productivity, Yeow and Sen (2004) presented a study in circuit board manufacturing that indicated $288,249 annual loss due to defcient inspection workspace and process. Given the safety and productivity implications, research has char- acterized the visual inspection process and to identify factors impact- ing performance. The most notable work is by Drury, 2001 , who have described the inspection process involving fve steps – (1) setup, (2) present, (3) search, (4) decide, and (5) respond – and validated two-stage model of search and decide for visual inspection (Drury, 1975; Spitz and Drury, 1978). The literature also contains research on performance impacts due to a wide range of factors, which are typ- ically classifed as task (Schoonard and Gould, 1973; Mital et al., 1998), environment (e.g., illumination (Wei and Konz, 1978),), individ- ual (e.g., age (Czaja and Drury, 1981; Harris, 1969; Megaw, 1979; Johnson and Funke, 1980), organizational (e.g., training (Gramopad- hye et al., 1997), feedback (Drury and Addison, 1973)), and social (e.g., communication (McCornck 1961)). See (2012) presents a com- prehensive literature review on visual inspection. Visual inspection research continues to accrue on this strong theo- retical and empirical foundation as inspection work evolves. Amongst the recent investigation is the use of automation or computer technol- ogy to form hybrid inspection systems (Kopardekar et al., 1993). Despite technological advances, complete automation of visual inspec- tion is not always feasible (Mital et al., 1998; Yang and Marefat, 1994). Manual inspection remains necessary in manufacturing of highly fexible designs when product samples are inadequate for training ma- chine vision algorithms, and when human expertise is required for trou- bleshooting misdetections or false alarms from sensor-based inspection systems (Chen et al., 2016). Thus, most inspection systems are in- creasingly setup to involve both human and automation, capitalizing on human expertise and machine capabilities (e.g., no fatigue thereby achieving 100% inspection). Jiang et al. (2003) conducted an empir- ical experiment comparing performance of human, automated and hy- brid inspection of circuit board defects, revealing that hybrid inspection can outperform human or automation alone, and risky systems (with relatively high false alarm rate) likely yield best accuracy. However, human trust in computerized inspection support systems tends to de- crease as false alarm increases (Jiang et al., 2004), indicating that careful design consideration is necessary to prevent disuse of inspection systems. Research on hybrid inspection systems in the literature also includes functional allocation between human and machine (Hou et ∗ Corresponding author. E-mail address: nathan.lau@vt.edu (N. Lau) https://doi.org/10.1016/j.apergo.2019.103022 Received 20 Febraury 2019; Received in revised form 19 October 2019; Accepted 28 November 2019 Available online xxx 0003-6870/© 2019.