Thursday, November 19, 2026
Science
No Result
View All Result
  • Login
  • HOME
  • SCIENCE NEWS
  • CONTACT US
  • HOME
  • SCIENCE NEWS
  • CONTACT US
No Result
View All Result
Scienmag
No Result
View All Result
Home Science News

Dual Neural Networks Guide Precise Laser Microprocessing in Superalloys

November 19, 2026
in Science News
Cassandra Pierce
By Cassandra Pierce Scienmag Editorial Profile - Systems Neuroscience
Reading Time: 5 mins read
0
Dual Neural Networks Guide Precise Laser Microprocessing in Superalloys

Dual Neural Networks Guide Precise Laser Microprocessing in Superalloys

65
SHARES
587
VIEWS
Share on FacebookShare on Twitter
ADVERTISEMENT

In high-precision manufacturing, managing laser interactions to generate nonpenetrating features such as blind holes, grooves, and stepped surfaces represents a substantial operational hurdle. While conventional industrial laser cutting focuses on through-hole penetration where energy beams simply clear the entire thickness of a workpiece, nonpenetrating geometries require micrometer-level depth precision without breaking through the bottom or eroding adjacent sidewalls. In a study published in Nanomanufacturing and Metrology, researchers Wenhao Han, Yunfeng Wang, and Wei Guo demonstrated a dual feedforward backpropagation neural network architecture capable of navigating the complex thermal dynamics of millisecond laser processing. By establishing a coupled prediction and simulation framework, the team achieved closed-loop control over microhole depth and diameter in Inconel 718, restricting processing discrepancies to an average error of less than five percent.

The engineering imperative for nonpenetrating microstructures spans aerospace propulsion, semiconductor architecture, and advanced energy hardware. Historically, machinists relied on orthogonal experimental arrays and empirical parameter fitting to approximate drilling depths. However, millisecond-duration pulsed lasers deliver high energy densities that trigger severe thermal accumulation, transient molten pool convection, dynamic vapor recoil, and plasma cloud generation. These multi-physical phenomena decouple laser energy parameters from material removal depth in an intensely nonlinear fashion, making manual trial-and-error procedures both labor-intensive and fundamentally incapable of delivering dependable depth control.

To overcome the limitations of traditional fitting models, the research team implemented a data-driven architecture that bifurcates the modeling workload into two specialized neural networks: an inverse prediction network and a forward simulation network. Because multiple distinct parameter combinations—such as varying pairings of peak power, pulse duration, and pulse counts—can produce identical hole depths and diameters, a single inverse model typically suffers from severe ambiguities and mathematical ill-posedness. The dual-model framework circumvents this instability by using the prediction model to propose operating parameters based on user-specified geometric dimensions, while the simulation model independently forecasts the geometric outcome of those proposed parameters, establishing a closed verification loop before executing physical laser drilling.

Physical experiments were conducted on a three-axis CNC motion platform equipped with a single-mode quasi-continuous wave fiber laser operating at a 1.08-micrometer wavelength. The laser system supported pulse durations ranging from 0.05 to 50 milliseconds, delivering a maximum average power of 1.2 kilowatts in pulsed operation, a peak power reaching 12 kilowatts, and a maximum duty ratio of 50 percent. The target workpiece consisted of a 6-millimeter-thick plate of Inconel 718, a high-temperature nickel-based superalloy widely utilized in aeronautical and energy machinery. Workpiece surfaces were cleaned in an ethanol bath inside an ultrasonic cleaner for five minutes prior to processing. To eliminate mechanical cross-talk and prevent spatter redeposition from distorting adjacent trials, drilling sites were spaced at 4-millimeter intervals, and experimental sequences were executed in randomized orders across five repeated runs per parameter set.

Non-energy processing conditions were rigorously standardized based on preliminary calibration tests. Argon shielding gas was introduced coaxially at a regulated pressure of 12 bar, chosen specifically to clear expelled molten metal without overcooling the molten pool or triggering chemical oxidation along the hole boundaries. The focal plane of the laser beam was set 1 millimeter above the top surface of the workpiece. Placing the beam waist above the metal deliberately lowered the peak energy density at the initial entrance plane, curtailing excessive surface melting and entrance spattering while maintaining a more uniform energy density profile as the beam diverged downward into the developing channel. The laser repetition frequency was held constant at 20 Hz, requiring pulse numbers to take integer values.

To investigate parameter coupling, a full-factorial experimental design was executed across three key energy variables: laser peak power, duty ratio, and pulse count. Peak power was adjusted across 13 distinct levels to capture a wide processing window, while the duty ratio and pulse counts were each varied across five discrete levels. Measurements demonstrated instrument depth deviations below two percent and diameter errors under one percent, though deep high-aspect-ratio holes exhibited minor optical edge-scattering artifacts near steep sidewall boundaries.

Building an accurate training dataset required systematic sample pruning to eliminate unrepresentative physical regimes. From the raw experimental trials, 324 samples were removed because excessive laser energy completely penetrated the 6-millimeter plate, precluding blind-hole depth quantification. An additional 135 samples were excluded as statistical outliers falling outside three standard deviations from their group means or displaying obvious physical anomalies, such as sidewall slag clogging that obstructed confocal optical access in low-energy shallow holes. The resulting dataset comprised 1,166 validated experimental samples spanning the full operational space, capturing the distribution of microhole depths and diameters under varying laser energy deliveries.

The neural network architectures were constructed within MATLAB’s deep learning environment using feedforward backpropagation algorithms trained via the Levenberg-Marquardt optimization technique. To equip the network against factory-floor disturbances—such as laser power jitter, microscopic alloy segregation, ambient temperature variations, and optical measurement noise—the researchers deliberately injected Gaussian noise with a 5 percent standard deviation into the inputs and outputs prior to training. Sensitivity testing revealed that a lower noise injection of 2.5 percent was largely filtered out by integer pulse rounding, offering minimal regularization, whereas a 10 percent noise level destroyed the network’s capacity to discern underlying physical patterns, causing the models to collapse into identical mean predictions.

Architectural optimization revealed a fundamental structural divergence between the two cooperating networks. The prediction model, tasked with resolving non-unique inverse parameter combinations from geometric targets, performed best using a neuron-sparse structure comprising four hidden layers configured with 3, 6, 9, and 12 neurons, trained on a subset of 812 data points covering duty ratios from one to three. In contrast, the deterministic simulation model, which maps defined energy parameters directly to unique physical geometries via governing thermal conduction dynamics, utilized a three-layer hidden architecture containing 7, 13, and 19 neurons trained across all 1,166 samples. Both networks incorporated L2 regularization to suppress overfitting, with penalty factors set to 0.001 for the prediction network and 0.01 for the simulation network to balance multidimensional parameter fitting.

Regression evaluations conducted on reserved testing datasets demonstrated high predictive fidelity across both networks. In standalone testing, the prediction model achieved a correlation coefficient (R-value) of 0.8684 when forecasting the primary governing variable, laser peak power. The simulation model achieved R-values of 0.9199 for hole depth and 0.8717 for diameter against experimental test sets. When operating as an integrated closed-loop unit—feeding target geometric features through the prediction model and routing its predicted process parameters directly into the simulation model—the combined system attained R-values of 0.9478 for hole depth and 0.9240 for hole diameter, confirming that the coupled framework effectively dampened error propagation.

Direct experimental verification confirmed the engineering viability of the dual-network methodology across ten physical test samples distributed throughout the applicable processing domain. Target geometric specifications were supplied to the prediction model to generate laser operating parameters, which were subsequently executed on the Inconel 718 plates. The closed-loop simulation predicted depth with an average error of 3.96 percent and diameter with an average error of 3.77 percent. Upon physical drilling, confocal microscope evaluations revealed an actual average depth error of 4.52 percent and an actual average diameter error of 4.59 percent, successfully validating the target threshold of plus or minus five percent control accuracy.

The researchers attributed the remaining variance between simulated predictions and physical outcomes to intrinsic thermo-fluid complexities and parameter discretization. During millisecond laser pulses, transient molten pools undergo turbulent convection and liquid splatter governed by stochastic balances among surface tension, gravity, and vapor recoil pressures. Furthermore, dynamic plasma clouds oscillate in density and temperature, intermittently shielding the metal surface and altering effective energy absorption. Finally, because pulse numbers must be executed as discrete integers, rounding pulse counts introduces minor step-wise energy offsets that cannot be infinitely tuned, establishing a physical baseline error that the dual neural network framework nevertheless contained within strict engineering tolerances.

Subject of Research: Closed-loop neural network control of millisecond laser blind-hole microprocessing in Inconel 718

Article Title: Experimental and Numerical Study of Millisecond Laser Micro-Processing Based on a Dual Neural Network System

Article References: Han, W., Wang, Y., & Guo, W. (2026). Experimental and Numerical Study of Millisecond Laser Micro-Processing Based on a Dual Neural Network System. Nanomanufacturing and Metrology, 9(1), Article 4. https://doi.org/10.1007/s41871-025-00287-4

Image Credits: AI Generated

DOI: 10.1007/s41871-025-00287-4

Keywords: laser microprocessing, neural networks, Inconel 718, nickel-based superalloys, blind-hole drilling, millisecond laser, quasi-continuous wave fiber laser, feedforward backpropagation, closed-loop simulation, confocal microscopy, thermal accumulation, process parameter optimization, Levenberg-Marquardt algorithm, L2 regularization, molten pool dynamics, precision manufacturing, nonpenetrating microstructures

Cite Scienmag News

Cassandra Pierce. (November 19, 2026). Dual Neural Networks Guide Precise Laser Microprocessing in Superalloys. Scienmag. https://scienmag.com/dual-neural-networks-guide-precise-laser-microprocessing-in-superalloys/

Cassandra Pierce. "Dual Neural Networks Guide Precise Laser Microprocessing in Superalloys." Scienmag, 19 November 2026, https://scienmag.com/dual-neural-networks-guide-precise-laser-microprocessing-in-superalloys/. Accessed 19 November 2026.

Cassandra Pierce. "Dual Neural Networks Guide Precise Laser Microprocessing in Superalloys." Scienmag. November 19, 2026. https://scienmag.com/dual-neural-networks-guide-precise-laser-microprocessing-in-superalloys/

Tags: blind-hole drillingclosed-loop simulationconfocal microscopyfeedforward backpropagationInconel 718L2-regularizationlaser microprocessingLevenberg-Marquardt algorithmmillisecond lasermolten pool dynamicsneural networksnickel-based superalloysnonpenetrating microstructuresprecision manufacturingProcess parameter optimizationquasi-continuous wave fiber laserthermal accumulation
Share26Tweet16
Previous Post

Bedside Electroencephalography Detects Nonconvulsive Status Epilepticus in Pediatric Emergency Patients

Next Post

Review Evaluates Intranasal Delivery Strategies Across Molecular and Cellular CNS Therapeutics

Related Posts

Marula Oil and Fulvic Acid Match Antibiotic Growth Promoters in Broilers
Science News

Marula Oil and Fulvic Acid Match Antibiotic Growth Promoters in Broilers

November 19, 2026
Huatanqushihuoxue Formula Attenuates Liver Cancer Progression in Rat Model
Science News

Huatanqushihuoxue Formula Attenuates Liver Cancer Progression in Rat Model

November 19, 2026
Pediatric Cannabinoid Intoxications in Palermo Hospital Primarily Affected Children Under Two
Science News

Pediatric Cannabinoid Intoxications in Palermo Hospital Primarily Affected Children Under Two

November 19, 2026
Bronchoscopic Amphotericin B Salvage Therapy Shows Acceptable Safety in Fungal Infections
Medicine

Bronchoscopic Amphotericin B Salvage Therapy Shows Acceptable Safety in Fungal Infections

November 19, 2026
Five Key Predictors Explain Seventy Percent of High-Rise Seismic Acceleration
Technology and Engineering

Five Key Predictors Explain Seventy Percent of High-Rise Seismic Acceleration

November 19, 2026
Waste Aluminium Foil Converted Into High-Purity Water Treatment Alum
Earth Science

Waste Aluminium Foil Converted Into High-Purity Water Treatment Alum

November 19, 2026
Next Post
Review Evaluates Intranasal Delivery Strategies Across Molecular and Cellular CNS Therapeutics

Review Evaluates Intranasal Delivery Strategies Across Molecular and Cellular CNS Therapeutics

  • Mothers who receive childcare support from maternal grandparents show more optimized

    Mothers who receive childcare support from maternal grandparents show more parental warmth, finds NTU Singapore study

    27656 shares
    Share 11059 Tweet 6912
  • University of Seville Breaks 120-Year-Old Mystery, Revises a Key Einstein Concept

    1061 shares
    Share 424 Tweet 265
  • Bee body mass, pathogens and local climate influence heat tolerance

    682 shares
    Share 273 Tweet 171
  • Researchers record first-ever images and data of a shark experiencing a boat strike

    546 shares
    Share 218 Tweet 137
  • Groundbreaking Clinical Trial Reveals Lubiprostone Enhances Kidney Function

    531 shares
    Share 212 Tweet 133
Science

Embark on a thrilling journey of discovery with Scienmag.com—your ultimate source for cutting-edge breakthroughs. Immerse yourself in a world where curiosity knows no limits and tomorrow’s possibilities become today’s reality!

RECENT NEWS

  • Marula Oil and Fulvic Acid Match Antibiotic Growth Promoters in Broilers
  • Huatanqushihuoxue Formula Attenuates Liver Cancer Progression in Rat Model
  • Pediatric Cannabinoid Intoxications in Palermo Hospital Primarily Affected Children Under Two
  • Bronchoscopic Amphotericin B Salvage Therapy Shows Acceptable Safety in Fungal Infections

Categories

  • Agriculture
  • Anthropology
  • Archaeology
  • Athmospheric
  • Biology
  • Biotechnology
  • Blog
  • Bussines
  • Cancer
  • Chemistry
  • Climate
  • Earth Science
  • Editorial Policy
  • Marine
  • Mathematics
  • Medicine
  • Pediatry
  • Policy
  • Psychology & Psychiatry
  • Science Education
  • Science News
  • Social Science
  • Space
  • Technology and Engineering

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 5,151 other subscribers

© 2026 Scienmag - Science Magazine

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • HOME
  • SCIENCE NEWS
  • CONTACT US

© 2026 Scienmag - Science Magazine

Discover more from Science

Subscribe now to keep reading and get access to the full archive.

Continue reading