Research Discovery & Dissemination for Industry Impact
BluSkyi gets your research in front of industry practitioners who can apply it.
BluSkyi & Researchers
On BluSkyi, industry practitioners sign up to receive a personalised weekly briefing of curated academic research. Each briefing includes BluSkyi Insight Articles, which are industry-focused summaries of recent and trending high-quality research work, plus a link to the full paper or preprint.
Early Access opens with engineering, with other sectors to follow.
Inside our six-stage Insight Article production process →Your Research on BluSkyi
We monitor new publications, along with earlier work that's rapidly gaining momentum, and select papers that meet our quality and industry-relevance criteria. Because of this, an older paper of yours could be featured too.
BluSkyi does not accept submissions directly from research teams or universities. Selection by our quality gate is what keeps the briefing trustworthy for subscribers.
When a paper or preprint is featured, we initiate a verification process with its listed corresponding author.
How verification works
- 1
Your paper is featured.
Your paper met our quality and industry-relevance criteria and has been summarised as an Insight Article.
- 2
We contact you.
The first time your work is featured, we email the paper's corresponding author with a link to review the Insight Article, and again for each new article if you've opted in to hear from us.
- 3
You verify.
Each Insight Article takes about 4 minutes to read. Approve it as is, or suggest corrections and clarifications. Once verified, the article carries a “Researcher Verified” tag.
What you get when you verify
Industry reach reporting
See how many practitioners engaged with your work, and where.
Greater public visibility
Subscriber briefings already link to your full paper. Verified articles are more likely to be selected for open, public features on our website and social channels, which supports your visibility and altmetrics.
Evidence for impact statements
Showcase your Insight Articles and reach metrics in funding applications and impact reporting.
Permission to reuse
Share your verified Insight Article on your personal, research group, department, or university websites and social media channels.
Featured Insight Articles
A sample of the research landing in our subscribers' briefings.
Ionic Liquid Additives Lose Their Anti-Wear Edge Once Oil Temperatures Rise
Phosphonium ionic liquids are increasingly seen as candidates to replace some conventional lubricant additives. This study tempers that expectation. On steel ball-on-disc contacts at 18 °C, one candidate, [P4442][DEP], delivered the lowest wear of any additive tested. At 60 °C and 80 °C none of the three liquids protected as well as zinc dialkyldithiophosphate (ZDDP). Nanoscale tests point to the reason: their protective films grow slowly and appear to need fresh wear to start.
Pilot Links Membrane CO2 Capture to Microbes That Make Grid-Quality Methane
The chain works, with conditions attached. A two-stage polyimide membrane unit reached 94.3% CO2 purity at 71% recovery, but only from a 27% CO2 feed; at 15% it managed 70.4%. Downstream, trickle-bed reactors fed that CO2 with electrolytic hydrogen held methane above 95% at a two-hour gas retention time. The authors estimate pilot capture cost at 150–200 €/t CO2, with compression electricity the main driver. Richer CO2 sources suit the membrane stage best.
Measuring Deflection, Not Torque, Sets Solid-State Battery Pressure to Within 3%
Sulfide-based solid-state batteries need tens of megapascals of stack pressure, yet a common approach is to tighten bolts and neglect the fixture's elastic relaxation. This study shows why that fails at scale and offers a cheap fix: compensate for the fixture's own elastic deflection and measure it directly. On 96 cm² pouch-cell fixtures the method held pressure error to 2.16% across 10–30 MPa, giving large-format cell research a documented, reproducible pressure-setting method.
Four Safety Gates Stand Between a Civil Turboprop and the Forward Air Control Role
A light civil turboprop can do much of what an airborne forward air controller needs: it climbs hard, loiters for hours and has well-damped dynamics that forgive a pilot looking outside the cockpit. But a single exit door, a sidestick that fouls the cockpit wall, weak pitch-force cues and no secure radio are hard gates, not trade-offs. For buyers of commercial derivative aircraft, the lesson is that civil airworthiness says little about military suitability until someone flies the mission.
Aviation's Tool Qualification Rules Score Highest for Transfer to Autonomous Driving
Autonomous driving is pushing vehicle software into territory where automotive safety processes look thin beside aviation's. Walking both industries along the V-model, the authors identify ten transferable practices and score nine for feasibility. DO-330-style tool qualification scores 9.42 out of 10, far ahead of the rest; requirements practices follow. Making authority involvement mandatory is the only proposal to score below the neutral mark of 5. The weightings come from one manufacturer, so the ranking is indicative.
Three-Path Load Sharing Makes a Hybrid Bridge Joint Simpler to Design
Hybrid girder bridges that splice a steel-concrete composite segment into a prestressed concrete span address the pavement and fatigue problems of steel box decks, but the joint between the two has lacked a clear design basis. This modelling study shows axial force split almost evenly between bearing plate and shear connectors, identifies points of diminishing return for plate thickness and joint length, and isolates a single connector row that governs shear design.
Deep Learning Surrogates Evaluate Loss-of-Coolant Scenarios 10,000 Times Faster Than Code
Deep learning surrogates trained on 1,020 stored thermal-hydraulic simulations of a large-break loss-of-coolant accident predict peak cladding temperature to within about 10 K in blowdown and 22 K in reflood, and evaluate roughly 10,000 times faster than the system code. They reproduce the code's sensitivity rankings, and their 95/95 upper bounds converge with the code's, though the authors caution that the surrogates tend to under-predict the high temperatures that set that limit. Licensing-grade confidence has not been reached.
Machine Learning Puts Atom-Scale Defect Fabrication in 2D Materials on Autopilot
Engineering materials atom by atom has been a hand-operated craft. This team has turned it into a closed loop: machine learning labels each atom in a live microscope image, a rules engine picks which sulfur atoms to remove, and a programmable controller fires the beam. The system seeds, places and grows metallic nanowire edges in monolayer MoS2 unaided; its directional-growth strategy succeeded in 65% of 42 runs, using models trained on a few dozen hand-labelled examples.
Implant Scaffolds Should Dissolve on the Body's Healing Schedule, Not the Chemist's
A degradable scaffold succeeds only if it disappears at the pace new tissue arrives. This review collates published degradation data for PLA, PVA, PCL and PU and lines them up against tissue healing times: PVA suits fast-healing skin and cornea, while PLA, PCL and PU fit slower bone, cartilage, ligament, neural and vascular repair. The rule is useful. The underlying numbers, drawn from mismatched test conditions and extrapolated, are not yet design data.
Robo-Taxi Fleet Controller Nears Oracle Profit and Adapts to New Cities in 10 Episodes
Robo-taxi services only pay off if idle vehicles are in the right place before demand appears. This work shows that a hybrid controller, with optimisation handling the hard constraints and a graph neural network deciding where vehicles should be, earns within 1.6–2.3% of a perfect-foresight oracle in simulation, transfers across cities, network layouts and simulator fidelities with modest losses, and adapts to a new city with 10 training episodes rather than 10,000.
Off-the-Shelf Machine Learning Predicts Deepwater Standpipe Pressure Without Custom Code
Machine learning for drilling hydraulics has largely stayed with data specialists, because the strongest hybrid models demand bespoke code. This study shows that four standard tree-ensemble models, run inside a no-code platform, can predict standpipe pressure across four deepwater gas wells with test R² between 0.968 and 0.994. The workflow is the real contribution. The accuracy still needs tougher validation, and every new field will need its own retraining.
A 155-Fold Faster Model Pulls Die Casting Vent Design Out of Trial and Error
The vents that let air escape a high-pressure die casting are still largely designed by prototype, because a single full 3D airflow simulation can take weeks. A coupled 0D–1D–3D model with adaptive time-stepping now runs nearly 155 times faster, finishing in hours on a standard PC while tracking pressures measured on an industrial test rig. The caveat is that its five fitted constants are proven only within a narrow envelope of vent sizes and tank volumes.
Adding Tropoelastin Gives Silk Hydrogels Micropores and Curbs Their Stiffening
Silk hydrogels suit light-based fabrication but carry two defects: pores too small for cells, and steady stiffening after they are made. Blending silk with tropoelastin in equal parts and curing with visible light addresses both in a single step, with no chemical modification. Pores grow roughly fivefold, four-week stiffening drops from about 12.9-fold to no more than 2.3-fold, and cells colonise implants in mice. Tissue sealing and printing remain proof-of-concept demonstrations.
Real Ocean Winds Undercut the Design Assumptions Behind Wind-Assisted Ship Propulsion
Wind-assisted propulsion is designed and tested against idealised wind. Two lidars on a chemical tanker show the real apparent wind is reshaped by hull, superstructure and the sails themselves, rarely follows the standard 1/7 or 1/9 power-law profiles, and carries strong fluctuations far more often than Gaussian models allow. For sail suppliers, ship operators and rule-makers, the inflow assumptions used in design, simulation and operation need revisiting, and the authors have released reference time series to do it.
Listening Model Cuts Fault-Detection Errors by More Than Half on Noisy Pump Audio
Acoustic monitoring promises non-contact fault detection for pumps, motors and power plant, but single-network models struggle with noise and scarce fault examples. DBAN-Net runs a convolutional branch and a recurrent branch in parallel and lets a lightweight attention gate weight them. On a public pump dataset it cut classification errors from 4.2% to 1.74% against the best baseline, and led at every training-data scale tested. Validation beyond pumps is still to come.
Ionic Liquid Additives Lose Their Anti-Wear Edge Once Oil Temperatures Rise
Phosphonium ionic liquids are increasingly seen as candidates to replace some conventional lubricant additives. This study tempers that expectation. On steel ball-on-disc contacts at 18 °C, one candidate, [P4442][DEP], delivered the lowest wear of any additive tested. At 60 °C and 80 °C none of the three liquids protected as well as zinc dialkyldithiophosphate (ZDDP). Nanoscale tests point to the reason: their protective films grow slowly and appear to need fresh wear to start.
Pilot Links Membrane CO2 Capture to Microbes That Make Grid-Quality Methane
The chain works, with conditions attached. A two-stage polyimide membrane unit reached 94.3% CO2 purity at 71% recovery, but only from a 27% CO2 feed; at 15% it managed 70.4%. Downstream, trickle-bed reactors fed that CO2 with electrolytic hydrogen held methane above 95% at a two-hour gas retention time. The authors estimate pilot capture cost at 150–200 €/t CO2, with compression electricity the main driver. Richer CO2 sources suit the membrane stage best.
Measuring Deflection, Not Torque, Sets Solid-State Battery Pressure to Within 3%
Sulfide-based solid-state batteries need tens of megapascals of stack pressure, yet a common approach is to tighten bolts and neglect the fixture's elastic relaxation. This study shows why that fails at scale and offers a cheap fix: compensate for the fixture's own elastic deflection and measure it directly. On 96 cm² pouch-cell fixtures the method held pressure error to 2.16% across 10–30 MPa, giving large-format cell research a documented, reproducible pressure-setting method.
Four Safety Gates Stand Between a Civil Turboprop and the Forward Air Control Role
A light civil turboprop can do much of what an airborne forward air controller needs: it climbs hard, loiters for hours and has well-damped dynamics that forgive a pilot looking outside the cockpit. But a single exit door, a sidestick that fouls the cockpit wall, weak pitch-force cues and no secure radio are hard gates, not trade-offs. For buyers of commercial derivative aircraft, the lesson is that civil airworthiness says little about military suitability until someone flies the mission.
Aviation's Tool Qualification Rules Score Highest for Transfer to Autonomous Driving
Autonomous driving is pushing vehicle software into territory where automotive safety processes look thin beside aviation's. Walking both industries along the V-model, the authors identify ten transferable practices and score nine for feasibility. DO-330-style tool qualification scores 9.42 out of 10, far ahead of the rest; requirements practices follow. Making authority involvement mandatory is the only proposal to score below the neutral mark of 5. The weightings come from one manufacturer, so the ranking is indicative.
Three-Path Load Sharing Makes a Hybrid Bridge Joint Simpler to Design
Hybrid girder bridges that splice a steel-concrete composite segment into a prestressed concrete span address the pavement and fatigue problems of steel box decks, but the joint between the two has lacked a clear design basis. This modelling study shows axial force split almost evenly between bearing plate and shear connectors, identifies points of diminishing return for plate thickness and joint length, and isolates a single connector row that governs shear design.
Deep Learning Surrogates Evaluate Loss-of-Coolant Scenarios 10,000 Times Faster Than Code
Deep learning surrogates trained on 1,020 stored thermal-hydraulic simulations of a large-break loss-of-coolant accident predict peak cladding temperature to within about 10 K in blowdown and 22 K in reflood, and evaluate roughly 10,000 times faster than the system code. They reproduce the code's sensitivity rankings, and their 95/95 upper bounds converge with the code's, though the authors caution that the surrogates tend to under-predict the high temperatures that set that limit. Licensing-grade confidence has not been reached.
Machine Learning Puts Atom-Scale Defect Fabrication in 2D Materials on Autopilot
Engineering materials atom by atom has been a hand-operated craft. This team has turned it into a closed loop: machine learning labels each atom in a live microscope image, a rules engine picks which sulfur atoms to remove, and a programmable controller fires the beam. The system seeds, places and grows metallic nanowire edges in monolayer MoS2 unaided; its directional-growth strategy succeeded in 65% of 42 runs, using models trained on a few dozen hand-labelled examples.
Implant Scaffolds Should Dissolve on the Body's Healing Schedule, Not the Chemist's
A degradable scaffold succeeds only if it disappears at the pace new tissue arrives. This review collates published degradation data for PLA, PVA, PCL and PU and lines them up against tissue healing times: PVA suits fast-healing skin and cornea, while PLA, PCL and PU fit slower bone, cartilage, ligament, neural and vascular repair. The rule is useful. The underlying numbers, drawn from mismatched test conditions and extrapolated, are not yet design data.
Robo-Taxi Fleet Controller Nears Oracle Profit and Adapts to New Cities in 10 Episodes
Robo-taxi services only pay off if idle vehicles are in the right place before demand appears. This work shows that a hybrid controller, with optimisation handling the hard constraints and a graph neural network deciding where vehicles should be, earns within 1.6–2.3% of a perfect-foresight oracle in simulation, transfers across cities, network layouts and simulator fidelities with modest losses, and adapts to a new city with 10 training episodes rather than 10,000.
Off-the-Shelf Machine Learning Predicts Deepwater Standpipe Pressure Without Custom Code
Machine learning for drilling hydraulics has largely stayed with data specialists, because the strongest hybrid models demand bespoke code. This study shows that four standard tree-ensemble models, run inside a no-code platform, can predict standpipe pressure across four deepwater gas wells with test R² between 0.968 and 0.994. The workflow is the real contribution. The accuracy still needs tougher validation, and every new field will need its own retraining.
A 155-Fold Faster Model Pulls Die Casting Vent Design Out of Trial and Error
The vents that let air escape a high-pressure die casting are still largely designed by prototype, because a single full 3D airflow simulation can take weeks. A coupled 0D–1D–3D model with adaptive time-stepping now runs nearly 155 times faster, finishing in hours on a standard PC while tracking pressures measured on an industrial test rig. The caveat is that its five fitted constants are proven only within a narrow envelope of vent sizes and tank volumes.
Adding Tropoelastin Gives Silk Hydrogels Micropores and Curbs Their Stiffening
Silk hydrogels suit light-based fabrication but carry two defects: pores too small for cells, and steady stiffening after they are made. Blending silk with tropoelastin in equal parts and curing with visible light addresses both in a single step, with no chemical modification. Pores grow roughly fivefold, four-week stiffening drops from about 12.9-fold to no more than 2.3-fold, and cells colonise implants in mice. Tissue sealing and printing remain proof-of-concept demonstrations.
Real Ocean Winds Undercut the Design Assumptions Behind Wind-Assisted Ship Propulsion
Wind-assisted propulsion is designed and tested against idealised wind. Two lidars on a chemical tanker show the real apparent wind is reshaped by hull, superstructure and the sails themselves, rarely follows the standard 1/7 or 1/9 power-law profiles, and carries strong fluctuations far more often than Gaussian models allow. For sail suppliers, ship operators and rule-makers, the inflow assumptions used in design, simulation and operation need revisiting, and the authors have released reference time series to do it.
Listening Model Cuts Fault-Detection Errors by More Than Half on Noisy Pump Audio
Acoustic monitoring promises non-contact fault detection for pumps, motors and power plant, but single-network models struggle with noise and scarce fault examples. DBAN-Net runs a convolutional branch and a recurrent branch in parallel and lets a lightweight attention gate weight them. On a public pump dataset it cut classification errors from 4.2% to 1.74% against the best baseline, and led at every training-data scale tested. Validation beyond pumps is still to come.
Common questions
Don't see your question below or have thoughts to share? Write to us at hello@bluskyi.com.