Detecting Wing Delamination with Ultrasonic Waves and AI

Aircraft structural engineers use a combination of guided ultrasonic Lamb waves and artificial intelligence to identify hidden internal damage within carbon-fiber composite wings. While composites offer high strength-to-weight ratios, they are susceptible to delamination—an internal separation of layers caused by external impacts like bird strikes or hail—which remains invisible on the wing's surface. To detect this threat without dismantling the aircraft, engineers install networks of piezoelectric sensors that transmit and receive ultrasonic Lamb waves across the wing structure. Because analyzing the complex wave signals is difficult due to structural noise and environmental variables, artificial intelligence algorithms are employed to interpret the acoustic data, accurately pinpointing the location and severity of the internal delamination in real time.

The Challenge of Carbon-Fiber Delamination

Carbon-fiber reinforced polymers are made of multiple thin layers, or plies, bonded together with resin. When a composite wing experiences a blunt force impact, the surface may bounce back and appear completely undamaged. However, the shockwave traveling through the material can break the resin bonds between the internal layers. This hidden damage is known as Barely Visible Impact Damage (BVID) or delamination. Over time, flight vibrations and aerodynamic stress cause the delaminated area to expand, severely reducing the structural integrity of the wing and risking catastrophic failure if left undetected.

Generating Guided Ultrasonic Lamb Waves

To find internal separations, engineers utilize Structural Health Monitoring (SHM) systems built directly into the aircraft. The core component of this system is an array of permanently attached piezoelectric wafer active sensors (PWAS).

When an electrical voltage is applied to a piezoelectric sensor, it vibrates and generates high-frequency acoustic waves. In thin, plate-like structures such as an aircraft wing skin, these vibrations form guided ultrasonic waves known as Lamb waves. Unlike bulk ultrasonic waves that only scan a tiny point directly beneath a handheld probe, Lamb waves travel long distances laterally through the entire thickness of the composite wing.

Lamb waves propagate in two main modes:

  • Symmetric (\(S_0\)) mode: The wave compresses and stretches the material parallel to the wing surface.
  • Antisymmetric (\(A_0\)) mode: The wave bends the material perpendicular to the wing surface.

How Lamb Waves Interact with Damage

In a healthy, intact carbon-fiber wing, a Lamb wave sent from a transmitter sensor will travel smoothly through the structure and arrive at a receiver sensor with a predictable speed, amplitude, and wave shape.

When a Lamb wave encounters a pocket of delamination, the separated layers disrupt the wave's path. This interaction alters the wave in three specific ways:

  1. Wave Scattering: Part of the wave bounces off the edges of the delamination, reflecting back toward the transmitter and dispersing in different directions.
  2. Attenuation: The wave loses energy as it passes through the damaged zone, arriving at the receiving sensor with a significantly weaker amplitude.
  3. Mode Conversion: As the wave passes through the thinner, separated layers of the delamination, a Symmetric wave may suddenly convert into an Antisymmetric wave, altering its speed and arrival time.

The Role of Artificial Intelligence in Signal Processing

While the physics of Lamb waves are effective at interacting with damage, reading the resulting sensor data is extremely difficult for human engineers. Aircraft wings are complex structures containing ribs, spars, fasteners, and varying thicknesses, all of which naturally reflect and distort ultrasonic waves. Furthermore, changes in altitude, cabin pressure, and extreme in-flight temperatures also alter the speed of the waves, creating false alarms.

Artificial intelligence solves this problem by filtering out environmental noise and recognizing the exact acoustic signature of delamination. Engineers implement AI in the following steps:

  • Feature Extraction: The raw ultrasonic waveforms collected by the sensors are converted into visual time-frequency maps using mathematical techniques like Continuous Wavelet Transforms.
  • Machine Learning Training: Engineers train deep learning models, particularly Convolutional Neural Networks (CNNs), using thousands of datasets. These datasets include baseline waves from undamaged wings under various temperatures, as well as waves recorded from wings with known delamination sizes and locations.
  • Anomaly Detection: During flight or maintenance checks, the AI compares real-time sensor data against its training. It automatically cancels out harmless signal variations caused by cold temperatures or structural bolts.
  • Damage Localization and Sizing: When the AI detects wave scattering and mode conversions that match its training for delamination, it triangulates the time-of-flight data from multiple sensors. The system then outputs a precise digital map of the wing, highlighting the exact coordinates and surface area of the hidden impact damage for maintenance crews.