Overview / Usage

The idea behind emerged from a clear need in the manufacturing industry: the necessity to automate and optimize the process of tool maintenance. Traditionally, this process has been manual, time-consuming, and prone to human error.

Our project brought some obstacles, it required high computational power and the flexibility to operate across various hardware architectures. This requirement was critical because the project aimed to bring processing capabilities directly to the client's location, often involving diverse and unpredictable hardware environments.

The decision to enable localized processing was, first, because we could ensure enhanced security and privacy, enable a federated learning architecture structure platform, faster responses with real-time analysis and offer tailored processing power based on needs.

This brings the question about the diversity in hardware architectures, which presented another layer of complexity. Our initial solutions were primarily optimized for proprietary GPUs, which posed limitations with compatibility, upgrades and scalability costs.

Methodology / Approach

Approach / Building the Solution

The turning point in our quest to overcome the computational and architectural challenges came with the introduction of Intel's oneAPI tools. It represented a paradigm shift in how we approached our project's technicalities.

Intel oneAPI provided a cross-architecture code model to function across different types of hardware, an optimization of algorithms to maximize performance regardless of the underlying setup and an ease of portability.

The result was more robust and efficient stands as a testament to the power of innovative technology when applied to real-world challenges. For a closer look at an MVP public available version, visit Tool Detective Project.

Deep Dive: How Intel Tools Helped

Intel's oneAPI tools helped us create a system that's not only faster but also more efficient at predicting when metal cutting tools need to be replaced. At the core of our project, we combined several of Intel's top-notch tools.

We especially made good use of the Intel Extension for PyTorch*. This let us get the most out of PyTorch, making it work even better on Intel's hardware. A challenging part was getting neural networks to work on FPGAs (Field-Programmable Gate Arrays).

We will fine-tune these neural networks to fit perfectly with what FPGAs can do. This meant tweaking how they're structured and how they process data, so they work as efficiently as possible. By doing this, we manage to speed up our processing times and get more accurate results.

The next section describes the specific software and hardware combinations that made this possible.

Technologies Used

The Software + Hardware Combo

Software:

• Intel® oneAPI
• Intel® Extension for PyTorch*
• Intel® Deep Neural Networks Library (Intel® DNNL)

Hardware:

• NVIDIA A100 GPU
• Intel® FPGA

More Resources

• https://www.intel.com/content/www/us/en/developer/overview.html
• https://www.intel.com/content/www/us/en/developer/tools/oneapi/toolkits.html
• https://www.intel.com/content/www/us/en/developer/tools/oneapi/overview.html
• https://www.intel.com/content/www/us/en/developer/articles/technical/comparing-cpus-gpus-and-fpgas-for-oneapi.html
• https://www.intel.com/content/www/us/en/developer/topic-technology/artificial-intelligence/reference-kit.html

Repository

https://github.com/Tool-Detective

Other links

• Homepage