Dr. Sogand Hasanzadeh and Dr. Behzad Esmaeili joined our AI4AECO event in August 2021. They presented three main applications of AI in “personalizing safety training,” “detecting risk-compensatory behavior,” and  “monitoring hazards and identifying the root causes of accidents,” as summarized below.

Research 1: Personalizing Safety Training via Wearable Technology

Challenges: While OSHA training is essential, the industry needs more personalized safety training and should move away from the “One Size Fits All” style of training. For instance, safety managers can create personalized safety training programs based on workers\’ cognition.  

Approach & Findings: The research team investigated the impacts of four common variables that cause incidents:

1. Attention distribution: To track workers’ attention within construction sites, the research team asked workers to wear eye-tracking glasses while doing their construction tasks. Throughout their research, the team determined that workers’ personality, culture, age, and safety training skills could play a crucial role in workers’ cognitive failure and distribution of their attention across their surroundings.
2. Memory load (e.g., family issues, health concerns, or fight with a spouse): The research team manipulated the memory load of the workers to investigate the impacts of high-load vs. low-load memory on workers’ performance and hazard detections. The team found the workers with high memory load could not focus on their tasks and, as a result, would miss fall hazards 3.8 more times than workers with low memory load.
3. Change blindness (e.g., dynamic construction site): The research team investigated how the dynamic conditions of a construction site would impact workers’ performance and their situational awareness. The findings show more experienced workers could identify changes and hazards more accurately and their reaction time might be higher than younger workers.
4. Time pressure: The team also explored how tight deadlines would change workers’ behavior and cognitive state. The team highlighted that time pressure negatively impacts hazard identification and attentional distribution of workers. The high-stress levels created by the tight deadlines caused workers to show more unsafe behavior, such as sitting close to the roof edge or keeping the hammer too close to fingers.

The team identified that personalized safety training enables workers to get better at distributing their attention across the scene within construction sites. In addition, personalized training helps workers gain higher situational awareness and balance their attention by observing their surroundings and scanning ahead for potential hazards.

One size does not fit all. Besides OSHA training, the construction workers need interactive safety training tailored to individual behaviors and circumstances.

Personalizing Safety Training via Wearable Technology

Research 2: Detecting Risk-compensatory Behavior

Challenges: Studies show 17% of fall accidents occurred among workers who had all essential fall safety protections. But why?

Approach & Findings: The research team simulated workers’ behavior in a 3D cave environment by manipulating the level of protection for fall hazards: no fall protection, injury-reducing fall protection, and all fall protection. The team determined that providing more safety protections produced a sense of invulnerability for workers, ultimately increasing workers’ risk-taking by up to 55%. When workers perceive the situation as safe (lower than an individual’s risk tolerance), time pressure and cognitive demand disrupts attention and requires a more significant share of cognitive resources. The workers usually adjust their risk-taking behavior and actions based on the risks and benefits associated with the situation.

Providing safety equipment or putting more technology in place will improve the protection, but will also reduce the perceived level of risk.

Detecting Risk-compensatory Behavior

Research 3: Monitoring Hazards and Identify the Root Causes of Accidents

Challenges: Although frequent quality inspections of safety conditions on-site can be a leading indicator of safety performance, several factors make it difficult for safety managers to increase the number of safety inspections in high-rise buildings. First, there are a limited number of safety managers in each company that may be located in construction sites across the county. Second, the large size and vertical construction of high-rise buildings make frequent inspections difficult. Therefore, increasing safety inspection frequency and observing hard-to-reach areas would significantly improve safety performance in high-rise construction projects.

Approach & Findings: To reduce the rate of accidents on-site, the research team built AI-assisted technologies to enable generating real-time safety warnings:

• They used photos taken by drones to detect hazardous situations and alert workers. For example, the AI algorithm detects floors and the availability of guardrails, as well as their spacing, and informs the team when there is a risk.
• The team used machine learning techniques, like decision trees, to combine multiple external and internal factors to understand the root causes of fall accidents. Using AI technology, the team can automate the process of analyzing accident reports and also help mitigate future accidents.

AI helps safety managers to monitor hazards and identify the root causes of accidents.

Monitoring Hazards and Identifying the Root Causes of Accidents