AI-Based Manufacturing Solutions

Our interdisciplinary engineering research team is developing a low-cost, flexible, user-friendly, and secure solution to help manufacturing firms detect defects using artificial intelligence techniques.

Small and medium-sized manufacturers often face barriers, including affordability, that make traditional automation and inspection upgrades difficult to justify. The challenges include: 

• Potential long downtime: System changes and equipment upgrades can interrupt production.
• Reliance on experience: Inspection quality may depend heavily on individual worker expertise.
• Expensive upgrades: Conventional inspection equipment can require major capital investment.
• Data sensitivity: Manufacturers need practical AI solutions that protect proprietary information.

Demonstration: Joint Cap Defect Detection

Our demonstration project applies computer vision by using convolutional neural networks (CNNs) to distinguish between acceptable and defective joint-cap conditions. 

A convolutional neural network is the kind of deep learning that makes self-driving cars possible or helps find abnormalities in medical scans. It begins with three-dimensional visual images and applies a convolutional layer of filters (called kernels) that detect edges and create a 3-D matrix that can be analyzed. Where other computations are algorithm-based, CNNs depend on pattern-based deep learning. Deep learning requires less computing power and data storage, which is why self-driving cars are possible without relying on supercomputers the size of a basketball court.

Our demonstration uses an open-source deep learning platform called PyTorch to inspect a joint cap. Here's how it works:

• Image-based inspection: An AI model processes input images from the camera to classify the part’s condition.
• Edge deployment: Inference is performed locally on the NVIDIA platform for fast response.
• Preliminary results: The prototype demonstrates automated classification with a clear Helioseismic and Magnetic Imager (HMI) result.

What Makes Our Approach Affordable and Reliable?

Zero-Trust Payload Isolation: Encrypting data directly on the local device ensures sensitive operational files are fully protected before leaving the local device.

Open Lightweight Encryption: AES-256 header encryption (-mhe=on) provides robust metadata protection without requiring proprietary software or paid platform licenses.

Encrypted Transit Channels: End-to-end transport is secured via standard HTTPS/TLS protocols integrated into modern cloud infrastructure. 

Flexible Network Architecture: Designed to support segmented cellular connections (SIM) or static IP endpoint tunnels for air-gapped facility safety.

Our In-Lab Test & Validation Plan

• Payload Encryption: Compress sample inspection datasets into multi-part 7z archives using -mhe flags to verify complete filename and metadata masking.
• Secure Cloud Offload: Transmit encrypted archives to cloud storage over HTTPS/TLS encrypted channels.
• Transport Inspection: Inspect network payload traffic using protocol analyzers (Wireshark) to confirm full end-to-end transit protection.
• Decrypt & Verify: Download archives to an isolated secondary endpoint and verify strict password enforcement prior to extraction.