⚡ electrical engineering

Human-centric multi-objective optimisation of operator safety and productivity

This study employs multi-objective metaheuristic optimization to identify cutting parameters that balance operator safety (minimizing heart rate and perceived risk) with productivity (maximizing material removal rate) in milling operations, revealing that the Genetic Algorithm offers the best trade-off while feed rate is the primary driver of operator stress.

Original authors: Gregoire Tambwe MBANGU, George STILWELL, Yilei ZHANG, Zuzhen Ji, Dirk PONS

Published 2026-07-17

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Gregoire Tambwe MBANGU, George STILWELL, Yilei ZHANG, Zuzhen Ji, Dirk PONS

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are the captain of a high-speed race car. Your goal is to drive as fast as possible to win the race, but there's a catch: the faster you go, the more your heart pounds, your hands shake, and the more terrified you feel about crashing. In the world of making things, specifically cutting metal with giant spinning tools, factory workers face a similar dilemma. They want to be productive—chipping away as much metal as possible, as quickly as possible—but they also need to stay safe and calm. If the machine vibrates too much or gets too loud, the worker's heart rate spikes, and their sense of safety drops. This isn't just about feeling good; a stressed-out operator is a tired operator, and a tired operator makes mistakes. So, scientists are trying to solve a tricky puzzle: How do we tune the machine to go fast without making the human driver feel like they're about to pass out?

This is the story of a new study that tried to solve this puzzle using a special kind of "digital brain" called artificial intelligence. The researchers didn't just guess; they built a computer model that acts like a super-smart coach. This coach watches three things at once: how fast the metal is being removed (productivity), how fast the worker's heart is beating (physiological stress), and how scared the worker feels (subjective safety). The goal was to find the perfect "sweet spot" where the machine works hard, but the human stays calm and confident. To do this, they tested three different types of AI coaches, each with its own personality, to see which one could find the best balance.

The Experiment: A Dance of Metal and Mind

The researchers set up a real-life test in a workshop. They had 11 volunteers, all trained to use a milling machine, perform 27 different cutting tasks each. That's 297 total tasks! For every task, they changed three main settings: how fast the tool spins, how deep it cuts into the metal, and how fast it moves forward. While the workers did their job, the researchers watched them closely. They strapped smartwatches to the workers' wrists to track their heart rates, placed microphones to listen to the noise, and attached vibration sensors to the machine to feel its shaking. Afterward, the workers filled out a survey, rating how dangerous, nervous, or uncomfortable they felt.

The team then used a powerful statistical tool called "Robust Linear Modelling" to turn all this messy data into clear rules. Think of this like a detective connecting the dots: they figured out exactly how much the heart rate would jump if the feed rate increased, or how the vibration level changed when the cut got deeper. They discovered that the worker's heart rate was the biggest clue for how safe they felt. If the heart was racing, the worker felt unsafe. Interestingly, they found that while the speed of the tool made the machine louder, it was the depth of the cut and the feed rate (how fast the tool moved) that really stressed the worker's body.

The AI Showdown: Three Coaches, One Goal

Once they had the rules, the researchers let three different AI algorithms try to find the perfect machine settings. Imagine these three algorithms as three different types of coaches trying to tune a race car:

The Genetic Algorithm (GA): This coach is like a natural selection expert. It mixes and matches different settings, keeps the ones that work best, and throws out the bad ones, slowly evolving a perfect solution.
The Artificial Immune System (AIS): This coach is inspired by the human body's immune system. It creates many copies of good solutions and tweaks them slightly, like white blood cells hunting down a virus, to find the safest, most effective path.
The Ant Colony Optimization (ACO): This coach acts like a swarm of ants. The ants leave "scent trails" (digital signals) on good paths. If a path leads to a great result, more ants follow it, reinforcing the trail.

The Results: Who Won the Race?

When the dust settled, the three coaches had very different strategies and results.

The AIS Coach (The Balanced Winner):
The Artificial Immune System turned out to be the most well-rounded performer. It managed to keep the workers' heart rates the lowest of all, at just 78 beats per minute (bpm). At the same time, it kept the productivity high, removing about 460 mm³ of metal per minute. It found a path where the workers felt safe (with a risk score of 1.01) without slowing the machine down too much. The study suggests this is the most "reasonable" approach, balancing the human's well-being with the factory's output.

The GA Coach (The Safety First Approach):
The Genetic Algorithm was a bit more cautious. It prioritized keeping the workers safe above all else. It achieved a very low heart rate of 80 bpm and the lowest perceived risk score of 1.10. However, to keep things so safe, it had to accept a slightly lower productivity rate of 399.85 mm³/min. It's like a driver who refuses to speed up even a little bit to ensure they never feel nervous.

The ACO Coach (The Speed Demon):
The Ant Colony Optimization was the most aggressive. It managed to find the highest productivity, chipping away 470 mm³ of metal per minute. But there was a catch: the workers' heart rates were higher, reaching 82 bpm, and their perceived risk was higher too. The study found that this algorithm was a bit unstable; in some tests, it even suggested impossible settings (like negative metal removal), showing it wasn't reliable enough for safety-critical jobs.

The Big Takeaway

The study didn't just find one "magic button" to fix everything. Instead, it showed that there is no single perfect setting that makes everyone happy all the time. You have to choose your trade-off. If you want the absolute safest, most comfortable experience for the worker, you go with the GA settings. If you want the best balance of safety and speed, the AIS approach is the winner. If you just want to go as fast as possible and don't mind the workers feeling a bit more stressed, ACO might get you there, but it's risky.

The researchers also found something surprising about the machine itself. They discovered that the feed rate (how fast the tool moves) was the main thing that stressed the workers out, while increasing the depth of cut actually seemed to help reduce the physiological load in their specific tests. This is a bit counter-intuitive, like finding that taking a bigger step actually makes you feel less tired than taking many small, quick ones.

In the end, this paper proves that we can use smart computer models to design factories that care about the people working in them. By using these AI coaches, factory managers can tune their machines not just to make the most money, but to keep their workers' hearts beating at a comfortable pace and their minds feeling safe. It's a step toward a future where "smart manufacturing" truly means "smart for humans, too."

Technical Summary: Human-centric Multi-objective Optimisation of Operator Safety and Productivity

Problem Statement
In modern smart manufacturing, a critical challenge exists in balancing operator safety with production efficiency during milling operations. Traditional approaches often prioritize productivity metrics, such as Material Removal Rate (MRR), while neglecting the physiological and psychological stress imposed on operators by machine settings and environmental conditions. High levels of vibration, noise, and aggressive cutting parameters can elevate operator heart rate (HR) and induce subjective feelings of unsafety, potentially compromising performance and long-term health. There is a scarcity of empirical studies that simultaneously quantify physiological load (specifically HR) and subjective safety perception (Qi) under varied milling parameters, nor do many studies integrate these human-centric metrics into a unified multi-objective optimisation framework.

Methodology
This study developed an AI-driven framework to optimise milling parameters by balancing operator safety, physiological stress, and productivity. The research employed a mixed-methods approach combining experimental data collection, robust statistical modelling, and metaheuristic optimisation.

Experimental Design: A full factorial design ( $3 \times 3 \times 3$ ) was implemented involving 11 participants performing 27 distinct machining tasks each, resulting in 297 data points. The independent variables were cutting speed (500–1100 rpm), axial depth of cut (0.1–0.5 mm), and feed rate (125–275 mm/min). The workpiece material was mild steel (AISI 1018).
Data Collection: Physiological stress was monitored via heart rate (HR) using a consumer-grade photoplethysmography (PPG) smartwatch. Environmental stressors (vibration and sound) were measured using accelerometers and sound level meters. Subjective safety perception was captured through a custom-designed eight-item survey questionnaire (Qi), covering dimensions such as perceived danger, nervousness, discomfort, cognitive load, and reliance on Personal Protective Equipment (PPE).
Statistical Modelling: Robust Linear Modelling (RLM) with Huber's T norm and Iteratively Reweighted Least Squares (IRLS) was used to model the relationships between machining parameters, environmental factors, HR, and subjective safety scores. This approach was selected to handle outliers and non-Gaussian noise inherent in physiological and acoustic data.
Optimisation Framework: Three population-based metaheuristic algorithms were applied to identify optimal cutting parameters:
1. Genetic Algorithm (GA): Utilizing selection, crossover, and mutation.
2. Artificial Immune System (AIS): Utilizing cloning and mutation based on affinity.
3. Ant Colony Optimisation (ACO): Utilizing probabilistic, pheromone-guided search.
  The optimisation aimed to maximise MRR while minimising HR and the subjective safety risk index ( $Q_i$ ). A composite objective function with weighted coefficients ( $w_1=0.4$ for HR, $w_2=0.4$ for MRR, $w_3=0.2$ for $Q_i$ ) was used to evaluate trade-offs.

Key Results

Predictive Modelling: The RLM analysis revealed that axial depth of cut and feed rate are the primary drivers of increased heart rate, while vibration showed a significant negative correlation with HR (suggesting a complex interaction or adaptation effect). Conversely, cutting speed was the dominant predictor for sound levels. Crucially, the study found that HR is the primary determinant of subjective safety perception ( $Q_i$ ), with machining parameters having little direct influence on $Q_i$ once HR is accounted for.
Algorithm Performance:
- AIS (Artificial Immune System): Demonstrated the most balanced performance. It achieved the lowest physiological load (minimum HR of 78 bpm) and the lowest safety risk ( $Q_i = 1.01$ ) while maintaining high productivity (MRR $\approx$ 460 mm³/min). It yielded the highest combined safety-productivity score (0.75–0.95 depending on the evaluation metric).
- GA (Genetic Algorithm): Prioritised safety, achieving a low HR (80 bpm) and the lowest subjective risk ( $Q_i = 1.10$ ), but at the cost of moderate productivity compared to AIS. It provided the best deterministic trade-off in specific runs.
- ACO (Ant Colony Optimisation): Exhibited instability and produced non-physical predictions (negative MRR) in deterministic evaluations. While it achieved the highest peak MRR (approx. 470 mm³/min), it did so at the expense of higher HR (82 bpm) and elevated perceived risk ( $Q_i = 2.27$ ).
Parameter Influence: Feed rate was identified as the dominant control parameter affecting operator stress and subjective safety. Interestingly, all algorithms converged on an optimal axial depth of cut of 0.3 mm for the best combined performance.

Significance and Claims
The paper claims to offer a novel integration of human-centric safety metrics with productivity optimisation in milling operations using multi-objective metaheuristic optimisation. Its primary contributions are:

Empirical Quantification: Providing empirical data linking specific milling parameters to real-time physiological stress (HR) and subjective safety perception, addressing a gap in literature where such metrics are often studied in isolation.
Methodological Efficacy: Demonstrating the efficacy of Robust Linear Modelling (RLM) for predictive modelling in manufacturing environments characterized by high variability and outliers.
Algorithmic Comparison: Establishing that while different metaheuristics offer distinct trade-offs, the Artificial Immune System (AIS) provides a superior balance for human-centric manufacturing, effectively reducing physiological load without sacrificing significant productivity.
Framework for Smart Manufacturing: The study presents a unified framework where objective physiological data and subjective safety perceptions are combined to identify optimal machining conditions, supporting the transition toward "smart manufacturing" that prioritises both human well-being and operational efficiency.

The authors maintain a modest stance, noting that while the study identifies optimal parameters within the tested experimental bounds, the findings are specific to the defined milling conditions and the specific algorithms employed. The work highlights the necessity of multi-objective frameworks to navigate the inherent conflicts between safety and productivity rather than relying on single-criterion optimisation.

The Experiment: A Dance of Metal and Mind

The AI Showdown: Three Coaches, One Goal

The Results: Who Won the Race?

The Big Takeaway

More like this