Visual AI simplifies HSE management for chemical manufacturers

Chemical processing is one of the most stringently regulated industries in the world. It is also one of the most heavily regulated in terms of health, safety and environment (HSE) factors. Employees are exposed to explosives, gases, flammable liquids and toxic infectious and oxidizing substances. They also work within the proximity of pressure vehicles that can leak, fail and succumb to design defects and corrosion. These all pose a real threat to these workers that can lead to sickness, injury and even death.

In following regulation and compliance standards, chemical processors in the United States are obligated to maintain precise product information and quality management data about chemicals and other materials, production techniques, packaging and processes, as well as employee illnesses and injuries related to their manufacturing operations and effects on the environment. They must also track toxic release data to the Environmental Protection Agency.

Despite inherent innovations and advancements, the chemical processing industry has been considered by some “a digital laggard” as it grapples with regulatory burdens. According to intelligence firm ABI Research (ABI), chemical companies have historically lacked confidence in the potential gains of digital transformation — and the few chemical manufacturers that did recognize the benefits struggled to scale up beyond the pilot phase.

In the mid-2000s, the industry turned a corner. Prompted by the arrival of the Internet of Things (IoT) and 3D printing, smart manufacturing was born. According to the EY CEO Outlook Survey 2022, digital transformation is now the second-most prominent capital issue for chemical firms across the globe. ABI predicts the chemical manufacturing industry will spend US $4.4 billion on digital transformation technologies in 2023 and US $7.4 billion on the digitalization of plants by 2031.

Even with the adoption of digitization to manage decarbonization and automation, chemical processors find HSE environments challenging to control due to insufficient cross-functional collaboration, disparate systems that do not communicate and undefined and unstandardized business processes.

Current health and safety protocols focus on training, procedure compliance, worker behavior and the supervised working environment for preventing workplace incidents. HSE managers are tasked with measuring and treating symptoms, such as unsafe acts, near misses and incidents, which are problematic with poor communication, lack of information and limited resources. Despite concerted efforts to minimize workplace accidents, injuries and deaths, the National Safety Council reported more than 4.2 million work-related medically consulted injuries in 2021 and 4,472 preventable work deaths. These ongoing occurrences emphasize the flaws in reactive solutions that lack preventive controls.

To rectify this, chemical processors are turning to artificial intelligence (AI) to drive HSE initiatives.

Early computer vision methods focused on low-level image processing techniques explicitly for extracting features from an image, like color, textures, edges, corners and shapes, and then building higher-level tasks for object recognition and semantic understanding composed of the lower-level building blocks.

Modern visual AI techniques can be trained by safety professionals to identify leading indicators of hazards before an incident actually happens. Visual AI allows chemical manufacturers to identify and stop numerous risky behaviors, prevent incidents and build more robust controls that reduce operational risk. Visual AI integrates with existing CCTV cameras to identify and understand imagery and visual data and then act upon it by generating actionable analytics and real-time alerts.

Computer vision algorithms implement foundational tasks that interpret video inputs. They include:

• Image classification — Classifies the primary object in an image.
• Object localization — Determines the region of interest that contains the primary object in an image.
• Object recognition — Combines object localization and classification to identify all objects in an image.
• Object verification — Determines if an object exists in an image.
• Object detection — Determines the location of a specific object in an image.
• Semantic segmentation — Breaks down an image into its component objects.
• Object tracking — Determines/predicts the position of an object in a sequence of video frames.

There is a visual AI product on the market today that can scale to thousands of cameras in days, using widely available and inexpensive GPU nodes. It works with virtually any camera system installed on a site, whether CCTV, Pan-Tilt-Zoom (PTZ), drones and/or mobile devices, and consolidates all video feeds into a single interface. The visual AI technology monitors employee movements, gathers data via heatmaps and situational awareness data, specifies unsafe actions that should shut down a process, creates engineering controls and sends real-time alerts to managers to help HSE supervisors optimize toward a zero-incident workplace. It observes and addresses previously undetected warning signs before they spiral into urgent and costly workplace problems.

Visual AI reduces the burden on employees and workflows by detecting, recording and alerting workers before a safety event occurs. It improves visibility and accountability through reporting and quantifiable results. Visual AI addresses trends and specific HSE issues with reports that include trends, worker shifts, specific areas and individual facilities, and it uses the reports for targeted training programs and controls and reduce regulatory burdens by automating reporting. It also disrupts existing safety patterns by automating records, alerts and controls. Visual AI gives HSE managers the power to quickly and reliably address procedures and behaviors that cause risky situations, create a system of improvement, ease the safety burden on employees and simplify HSE program improvements.

With a low-code/no-code integration framework, which means no data scientist or programmer is required, visual AI deploys either on-premises or off and keeps data secure on-site or in the cloud. And its continuous-learning training module improves product performance over time to mitigate false positives and negatives.

Market demand and technological improvements are fueling transformational changes across the chemical processing industry. By 2023, more than 50 percent of all data analysis by deep neural networks will be at the point of capture in an edge system, a significant increase from less than five percent in 2019. By 2025, video analytics will be a standard element in two-thirds of new video surveillance installations, compared with less than 30 percent in 2020. By 2025, 80 percent of mass production facilities will employ advanced machine vision-based quality assurance (QA), compared with just five percent in 2021. And by 2030, the global computer vision market is projected to reach $41.11 billion, reflecting a 16 percent CAGR since 2020.

Chemical processors must put effort into improving the sustainability of health and safety performances in the workplace, with risks categorized as regulatory compliance risks, occupational hazards, natural disasters, employee safety hazards and environmental impacts. Manual, ad hoc, non-standardized HSE strategies and reporting are slow, insufficient and antiquated. Under this method, incidents are captured and recorded. Still, HSE managers fail to ingest and understand all the safety data and identify emerging risks and safety issues, which hinders mitigation strategies for preventing common issues. By adding visual AI to HSE management, chemical processors will find integrated data for one clear view of HSE across the entire organization.

Visual AI proves its efficiency in predictive-based safety by providing recommendations and forecasting potential failures or accidents based on the data it generates. Ultimately, this solution helps chemical processing companies detect and enforce safer operations and protect workers to the greatest degree possible. Major industries such as energy, government, retail, construction and mining are already taking advantage of visual AI technology for HSE initiatives. Chemical processors should do the same.

Jaidev Amrite is head of product for Visual AI Advisor and DeepNLP for SparkCognition.