Sepehr Makhsous
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Sepehr Makhsous has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Alexander V. Mamishev
Department of Electrical and Computer Engineering, University of Washington, Seattle, WA 98195, USA
5 shared publications - Igor Novosselov
University of Washington
4 shared publications - Zheng Liu
Department of Electrical and Computer Engineering, University of Washington, Seattle, WA 98195, USA
3 shared publications - Daniel C. N. Chan
Department of Restorative Dentistry, University of Washington, Seattle, WA, USA,
2 shared publications - Denise Wilson
Department of Electrical and Computer Engineering, University of Washington, Seattle, WA, USA
2 shared publications - Gokul Nathan
Department of Electrical and Computer Engineering, University of Washington, Seattle, WA 98195, USA
2 shared publications
Latest Publications
- Open Access
Firefly: A Low-Resolution Thermal Trigger for Gated Sensor Fusion in Human–Robot Collaborative Manufacturing
Automated guided vehicles and mobile robots are present on many Intelligent Manufacturing factory floors. Approximately 7.0 million U.S. workers also work on those same floors. During 2011 through 2017, forklifts caused 614 worker fatalities. Currently, no commercially available collision-avoidance sensor can manage the complexity associated with working within a plant. Therefore, fusing multiple modalities allows a system designer to gate the more expensive ones behind some minimal signal that indicates all systems should be active. Such minimal signals are expected to include body heat that even stationary personnel would have difficulty suppressing. In addition, the signal-capturing sensors ought to operate at milliwatt levels of power while keeping workers’ privacy. One type of device that has been found suitable for these requirements is a coarse thermopile array. Such a device can be fused with a rangefinder that determines distance to form a gated system. Yet the measurement capability frontier of the thermopile arrays in practical scenarios has not been explored. In addition, there is currently no figure of merit to quantify the trade-ff between induced errors and system-level power saving. In order to address these gaps, we built the fusion system, Firefly, together with a training-free detector and tested them against a human and a mannequin 1.2 to 7 m away in four thermally distinct environments, one of which is the floor of an operating factory. We found that the thermal array’s detection range varies with the environment but is predictable in a thermally uniform environment. Based on these experiments, we also proposed the Gated-Fusion Efficiency (GFE), a figure of merit that converts a trigger’s error rates into system-level power savings. Together, the capability boundary and the GFE help guide sensor-fusion system design in industrial manufacturing toward safer working environments.
Open AccessFirefly: A Low-Resolution Thermal Trigger for Gated Sensor Fusion in Human–Robot Collaborative Manufacturing
Automated guided vehicles and mobile robots are present on many Intelligent Manufacturing factory floors. Approximately 7.0 million U.S. workers also work on those same floors. During 2011 through 2017, forklifts caused 614 worker fatalities. Currently, no commercially available collision-avoidance sensor can manage the complexity associated with working within a plant. Therefore, fusing multiple modalities allows a system designer to gate the more expensive ones behind some minimal signal that indicates all systems should be active. Such minimal signals are expected to include body heat that even stationary personnel would have difficulty suppressing. In addition, the signal-capturing sensors ought to operate at milliwatt levels of power while keeping workers’ privacy. One type of device that has been found suitable for these requirements is a coarse thermopile array. Such a device can be fused with a rangefinder that determines distance to form a gated system. Yet the measurement capability frontier of the thermopile arrays in practical scenarios has not been explored. In addition, there is currently no figure of merit to quantify the trade-ff between induced errors and system-level power saving. In order to address these gaps, we built the fusion system, Firefly, together with a training-free detector and tested them against a human and a mannequin 1.2 to 7 m away in four thermally distinct environments, one of which is the floor of an operating factory. We found that the thermal array’s detection range varies with the environment but is predictable in a thermally uniform environment. Based on these experiments, we also proposed the Gated-Fusion Efficiency (GFE), a figure of merit that converts a trigger’s error rates into system-level power savings. Together, the capability boundary and the GFE help guide sensor-fusion system design in industrial manufacturing toward safer working environments.
- Open Access
AETOS: Near-Field Urban Pollution Monitoring Using a Distributed Sensor Network and Backward–Forward Lagrangian Modeling: A Fireworks Case Study at Lake Union, Seattle
Public fireworks shows are widely used for national holiday celebrations, religious ceremonies, and sports events. However, fireworks produce short-lived atmospheric particulate matter (PM) peaks, posing a risk to spectators’ health. Every year, public fireworks shows expose over 140 million Americans to episodic transient PM2.5 peaks that exceed air quality standards by up to 10 times in the United States alone. Yet the magnitude and timing of this representativeness gap have not been measured within an event. This paper investigates the Independence Day fireworks display over Lake Union, Seattle, WA, USA, using a low-cost PM network deployed up to 2 km from the launch site. We evaluated the network observations against routine monitoring data from regulatory stations within 10 km of the launch site and generated a backward–forward Lagrangian stochastic dispersion model, calibrated with the network. Spectator zone PM2.5 concentrations varied by over threefold across sensors within 1.5 km. Duration above the World Health Organization (WHO) 24 h PM2.5 guideline concentration level at network sensors varied from 1 min to nearly 30 min, and regulatory hourly averaging retained 57% of the near-field peak signal on average and only 22% in the worst case. Peak detection at some regulatory stations was delayed by one to two hours compared with the network. The dispersion model, using only network data and public regional wind data, captured the launch site location to within ~100 m and provided minute-scale estimates of PM2.5 and PM1 concentrations across the spectator zone. The findings demonstrate and quantify, for a single event, the representativeness gap expected when transient near-field fireworks plumes are evaluated using spatially sparse and hourly averaged regulatory observations.
Open AccessAETOS: Near-Field Urban Pollution Monitoring Using a Distributed Sensor Network and Backward–Forward Lagrangian Modeling: A Fireworks Case Study at Lake Union, Seattle
Public fireworks shows are widely used for national holiday celebrations, religious ceremonies, and sports events. However, fireworks produce short-lived atmospheric particulate matter (PM) peaks, posing a risk to spectators’ health. Every year, public fireworks shows expose over 140 million Americans to episodic transient PM2.5 peaks that exceed air quality standards by up to 10 times in the United States alone. Yet the magnitude and timing of this representativeness gap have not been measured within an event. This paper investigates the Independence Day fireworks display over Lake Union, Seattle, WA, USA, using a low-cost PM network deployed up to 2 km from the launch site. We evaluated the network observations against routine monitoring data from regulatory stations within 10 km of the launch site and generated a backward–forward Lagrangian stochastic dispersion model, calibrated with the network. Spectator zone PM2.5 concentrations varied by over threefold across sensors within 1.5 km. Duration above the World Health Organization (WHO) 24 h PM2.5 guideline concentration level at network sensors varied from 1 min to nearly 30 min, and regulatory hourly averaging retained 57% of the near-field peak signal on average and only 22% in the worst case. Peak detection at some regulatory stations was delayed by one to two hours compared with the network. The dispersion model, using only network data and public regional wind data, captured the launch site location to within ~100 m and provided minute-scale estimates of PM2.5 and PM1 concentrations across the spectator zone. The findings demonstrate and quantify, for a single event, the representativeness gap expected when transient near-field fireworks plumes are evaluated using spatially sparse and hourly averaged regulatory observations.
- Open Access
Before Smelling the Video: A Two-Stage Pipeline for Interpretable Video-to-Scent Plans
Olfactory cues can enhance immersion in interactive media, yet smell remains rare because it is difficult to author and synchronize with dynamic video. Prior olfactory interfaces rely on designer triggers and fixed event-to-odor mappings that do not scale to unconstrained content. This work examines whether semantic planning for smell is intelligible to people before physical scent delivery. We present a video-to-scent planning pipeline that separates visual semantic extraction using a vision-language model from semantic-to-olfactory inference using a large language model. Two survey studies compare system-generated scent plans with over-inclusive and naive baselines. Results show consistent preference for plans that prioritize perceptually salient cues and align scent changes with visible actions, supporting semantic planning as a foundation for future olfactory media systems.
Open AccessBefore Smelling the Video: A Two-Stage Pipeline for Interpretable Video-to-Scent Plans
Olfactory cues can enhance immersion in interactive media, yet smell remains rare because it is difficult to author and synchronize with dynamic video. Prior olfactory interfaces rely on designer triggers and fixed event-to-odor mappings that do not scale to unconstrained content. This work examines whether semantic planning for smell is intelligible to people before physical scent delivery. We present a video-to-scent planning pipeline that separates visual semantic extraction using a vision-language model from semantic-to-olfactory inference using a large language model. Two survey studies compare system-generated scent plans with over-inclusive and naive baselines. Results show consistent preference for plans that prioritize perceptually salient cues and align scent changes with visible actions, supporting semantic planning as a foundation for future olfactory media systems.
- 13 August 2025
- Published by Elsevier in American Journal of Orthodontics and Dentofacial Orthopedics
- Vol. 168 (5) , 563-578.e4
Mitigation of aerosol particles during composite attachment removal
- 13 August 2025
- Published by Elsevier in American Journal of Orthodontics and Dentofacial Orthopedics
- Vol. 168 (5) , 563-578.e4
Mitigation of aerosol particles during composite attachment removal
- Open Access
Quantitative Assessment of Soldering-Induced PM2.5 Exposure Using a Distributed Sensor Network in Instructional Laboratory Settings
Soldering is a common engineering practice that releases airborne particulate matter (PM), contributing to significant long-term respiratory risk. The health impact of this exposure is significant, with up to 22% of soldering workers worldwide being diagnosed with conditions such as occupational asthma, restrictive lung disease, and bronchial obstruction. Studies have reported that soldering can produce PM2.5 concentrations up to 10 times higher than the U.S. Environmental Protection Agency’s (EPA) 24 h exposure limit of 35.0 μg/m3—posing significant respiratory and cognitive health risks under chronic exposure. These hazards remain underappreciated by novice engineers in academic and entry-level industrial environments, where safety practices are often informal or inconsistently applied. Air purification systems offer a mitigation approach; however, performance varies significantly with model and placement, and independent validation is limited. This study uses an indoor air quality monitoring system consisting of six AeroSpec sensors to measure PM2.5–10 concentrations during soldering sessions conducted with and without commercial air purifiers. Tests were conducted with and without a selection of commercial air purifiers, and measurements were recorded under consistent spatial and temporal conditions. Datasets were analyzed to evaluate purifier effectiveness and the influence of placement on pollutant distribution. The findings provide independent validation of air purifier capabilities and offer evidence-based suggestions for minimizing particulate exposure and improving safety in laboratory soldering environments.
Open AccessQuantitative Assessment of Soldering-Induced PM2.5 Exposure Using a Distributed Sensor Network in Instructional Laboratory Settings
Soldering is a common engineering practice that releases airborne particulate matter (PM), contributing to significant long-term respiratory risk. The health impact of this exposure is significant, with up to 22% of soldering workers worldwide being diagnosed with conditions such as occupational asthma, restrictive lung disease, and bronchial obstruction. Studies have reported that soldering can produce PM2.5 concentrations up to 10 times higher than the U.S. Environmental Protection Agency’s (EPA) 24 h exposure limit of 35.0 μg/m3—posing significant respiratory and cognitive health risks under chronic exposure. These hazards remain underappreciated by novice engineers in academic and entry-level industrial environments, where safety practices are often informal or inconsistently applied. Air purification systems offer a mitigation approach; however, performance varies significantly with model and placement, and independent validation is limited. This study uses an indoor air quality monitoring system consisting of six AeroSpec sensors to measure PM2.5–10 concentrations during soldering sessions conducted with and without commercial air purifiers. Tests were conducted with and without a selection of commercial air purifiers, and measurements were recorded under consistent spatial and temporal conditions. Datasets were analyzed to evaluate purifier effectiveness and the influence of placement on pollutant distribution. The findings provide independent validation of air purifier capabilities and offer evidence-based suggestions for minimizing particulate exposure and improving safety in laboratory soldering environments.
- Open Access
Methodology for Addressing Infectious Aerosol Persistence in Real-Time Using Sensor Network
Human exposure to infectious aerosols results in the transmission of diseases such as influenza, tuberculosis, and COVID-19. Most dental procedures generate a significant number of aerosolized particles, increasing transmission risk in dental settings. Since the generation of aerosols in dentistry is unavoidable, many clinics have started using intervention strategies such as area-filtration units and extraoral evacuation equipment, especially under the relatively recent constraints of the pandemic. However, the effectiveness of these devices in dental operatories has not been studied. Therefore, the ability of dental personnel to efficiently position and operate such instruments is also limited. To address these challenges, we utilized a real-time sensor network for assessment of aerosol dynamics during dental restoration and cleaning producers with and without intervention. The strategies tested during the procedures were (i) local area High-Efficiency Particle Air (HEPA) filters and (ii) Extra-Oral Suction Device (EOSD). The study was conducted at the University of Washington School of Dentistry using a network of 13 fixed sensors positioned within the operatory and one wearable sensor worn by the dental operator. The sensor network provides time and space-resolved particulate matter (PM) data. Three-dimensional (3D) visualization informed aerosol persistence in the operatory. It was found that area filters did not improve the overall aerosol concentration in dental offices in a significant way. A decrease in PM concentration by an average of 16% was observed when EOSD equipment was used during the procedures. The combination of real-time sensors and 3D visualization can provide dental personnel and facility managers with actionable feedback to effectively assess aerosol transmission in medical settings and develop evidence-based intervention strategies.
Open AccessMethodology for Addressing Infectious Aerosol Persistence in Real-Time Using Sensor Network
Human exposure to infectious aerosols results in the transmission of diseases such as influenza, tuberculosis, and COVID-19. Most dental procedures generate a significant number of aerosolized particles, increasing transmission risk in dental settings. Since the generation of aerosols in dentistry is unavoidable, many clinics have started using intervention strategies such as area-filtration units and extraoral evacuation equipment, especially under the relatively recent constraints of the pandemic. However, the effectiveness of these devices in dental operatories has not been studied. Therefore, the ability of dental personnel to efficiently position and operate such instruments is also limited. To address these challenges, we utilized a real-time sensor network for assessment of aerosol dynamics during dental restoration and cleaning producers with and without intervention. The strategies tested during the procedures were (i) local area High-Efficiency Particle Air (HEPA) filters and (ii) Extra-Oral Suction Device (EOSD). The study was conducted at the University of Washington School of Dentistry using a network of 13 fixed sensors positioned within the operatory and one wearable sensor worn by the dental operator. The sensor network provides time and space-resolved particulate matter (PM) data. Three-dimensional (3D) visualization informed aerosol persistence in the operatory. It was found that area filters did not improve the overall aerosol concentration in dental offices in a significant way. A decrease in PM concentration by an average of 16% was observed when EOSD equipment was used during the procedures. The combination of real-time sensors and 3D visualization can provide dental personnel and facility managers with actionable feedback to effectively assess aerosol transmission in medical settings and develop evidence-based intervention strategies.
- Open Access
- 28 April 2021
- Published by MDPI AG in Preprints.org
Methodology for Addressing Infectious Aerosol Persistence in Real-Time Using Sensor Network
Human exposure to infectious aerosols results a transmission of diseases, such as influenza, tuberculosis, and COVID-19. Most dental procedures generate a significant number of aerosolized particles, increasing transmission risk in dental settings. Since the generation of aerosols in dentistry is unavoidable, many clinics started using intervention strategies such as area-filtration units and extraoral evacuation equipment, especially under the relatively recent constraints of the pandemics. However, the effectiveness of these devices in dental operatories has not been studied. Therefore, the dental personnel's ability to position and operate such instruments efficiently is also limited. To address these challenges, we utilized a real-time sensor network for assessment of aerosol dynamics during dental restoration and cleaning producers with and without intervention. The strategies tested during the procedures were (i) local area high-efficiency particle air (HEPA) filters and (ii) extra-oral suction device (EOSD). The study was conducted at the University of Washington School of Dentistry using a network of thirteen fixed sensors positioned within the operatory and one wearable sensor worn by the dental operator. The sensor network provides time and space-resolved particulate matter (PM) data. Three-dimensional (3D) visualization informs aerosol persistence in the operatory. It was found that area filters did not improve the overall aerosol concentration in dental offices significantly. An average of 16% decrease in PM concentration was observed when EOSD equipment was used during the procedures. The combination of real-time sensors and 3D visualization can provide dental personnel and facility mangers with actionable feedback to effectively assess aerosol transmission in medical settings and develop evidence-based intervention strategies.
- 28 April 2021
- Published by MDPI AG in Preprints.org
Open AccessMethodology for Addressing Infectious Aerosol Persistence in Real-Time Using Sensor Network
Human exposure to infectious aerosols results a transmission of diseases, such as influenza, tuberculosis, and COVID-19. Most dental procedures generate a significant number of aerosolized particles, increasing transmission risk in dental settings. Since the generation of aerosols in dentistry is unavoidable, many clinics started using intervention strategies such as area-filtration units and extraoral evacuation equipment, especially under the relatively recent constraints of the pandemics. However, the effectiveness of these devices in dental operatories has not been studied. Therefore, the dental personnel's ability to position and operate such instruments efficiently is also limited. To address these challenges, we utilized a real-time sensor network for assessment of aerosol dynamics during dental restoration and cleaning producers with and without intervention. The strategies tested during the procedures were (i) local area high-efficiency particle air (HEPA) filters and (ii) extra-oral suction device (EOSD). The study was conducted at the University of Washington School of Dentistry using a network of thirteen fixed sensors positioned within the operatory and one wearable sensor worn by the dental operator. The sensor network provides time and space-resolved particulate matter (PM) data. Three-dimensional (3D) visualization informs aerosol persistence in the operatory. It was found that area filters did not improve the overall aerosol concentration in dental offices significantly. An average of 16% decrease in PM concentration was observed when EOSD equipment was used during the procedures. The combination of real-time sensors and 3D visualization can provide dental personnel and facility mangers with actionable feedback to effectively assess aerosol transmission in medical settings and develop evidence-based intervention strategies.
- Open Access
DietSensor: Automatic Dietary Intake Measurement Using Mobile 3D Scanning Sensor for Diabetic Patients
Diabetes is a global epidemic that impacts millions of people every year. Enhanced dietary assessment techniques are critical for maintaining a healthy life for a diabetic patient. Moreover, hospitals must monitor their diabetic patients’ food intake to prescribe a certain amount of insulin. Malnutrition significantly increases patient mortality, the duration of the hospital stay, and, ultimately, medical costs. Currently, hospitals are not fully equipped to measure and track a patient’s nutritional intake, and the existing solutions require an extensive user input, which introduces a lot of human errors causing endocrinologists to overlook the measurement. This paper presents DietSensor, a wearable three-dimensional (3D) measurement system, which uses an over the counter 3D camera to assist the hospital personnel with measuring a patient’s nutritional intake. The structured environment of the hospital provides the opportunity to have access to the total nutritional data of any meal prepared in the kitchen as a cloud database. DietSensor uses the 3D scans and correlates them with the hospital kitchen database to calculate the exact consumed nutrition by the patient. The system was tested on twelve volunteers with no prior background or familiarity with the system. The overall calculated nutrition from the DietSensor phone application was compared with the outputs from the 24-h dietary recall (24HR) web application and MyFitnessPal phone application. The average absolute error on the collected data was 73%, 51%, and 33% for the 24HR, MyFitnessPal, and DietSensor systems, respectively.
Open AccessDietSensor: Automatic Dietary Intake Measurement Using Mobile 3D Scanning Sensor for Diabetic Patients
Diabetes is a global epidemic that impacts millions of people every year. Enhanced dietary assessment techniques are critical for maintaining a healthy life for a diabetic patient. Moreover, hospitals must monitor their diabetic patients’ food intake to prescribe a certain amount of insulin. Malnutrition significantly increases patient mortality, the duration of the hospital stay, and, ultimately, medical costs. Currently, hospitals are not fully equipped to measure and track a patient’s nutritional intake, and the existing solutions require an extensive user input, which introduces a lot of human errors causing endocrinologists to overlook the measurement. This paper presents DietSensor, a wearable three-dimensional (3D) measurement system, which uses an over the counter 3D camera to assist the hospital personnel with measuring a patient’s nutritional intake. The structured environment of the hospital provides the opportunity to have access to the total nutritional data of any meal prepared in the kitchen as a cloud database. DietSensor uses the 3D scans and correlates them with the hospital kitchen database to calculate the exact consumed nutrition by the patient. The system was tested on twelve volunteers with no prior background or familiarity with the system. The overall calculated nutrition from the DietSensor phone application was compared with the outputs from the 24-h dietary recall (24HR) web application and MyFitnessPal phone application. The average absolute error on the collected data was 73%, 51%, and 33% for the 24HR, MyFitnessPal, and DietSensor systems, respectively.