In-depth digital twin developed by complex artificial intelligence, enabling 3D printing to enter production

Today, we are accustomed to seeing AI take over more and more tasks-not only in our daily lives, but also in medical applications or industrial production. The development of artificial intelligence has made great progress. It is now possible to use artificial intelligence to predict component failures in production or extract information from images to perform interference tasks within a fraction of a second.

This issue, combined with3D printingIn the field, how to promote the development of complex artificial intelligence deep digital twins3D printingEntering production, 3D Science Valley and Gu You will come to appreciate the ongoing new atmosphere of artificial intelligence empowered additive manufacturing.

Nnaisense

The stability of AI empowered processing

First of all, let’s understand, why can’t achieve good quality control for the large amount of monitoring data collected in additive manufacturing (AM) machines? We need to manufacture a typical additive manufacturing process consisting of thousands of layers, and each layer will generate several MB of rasterization or time series monitoring and sensor data to generate technical data packages for common production scenarios. In terms of time saving, repeatability and data efficiency, the benefits of using data to guide processing operations will be huge.

So far, classical data analysis using image and signal processing techniques can also work, in fact they are widely used for this task. However, because the different part geometries and process conditions in AM additive manufacturing equipment can cause so many process interdependencies, it turns out that it is almost impossible to find the correct parameterization of traditional analysis algorithms to provide “correct” instructions to the equipment of.

In the book “Design for Additive Manufacturing (DfAM) Guide”, the Ishikawa diagram of the factors affecting the quality of AM parts is cited. In the Ishikawa diagram, more than 160 factors that affect the quality of processing are listed in detail. It is only the laser scanning process. , Including scanning line length, scanning line type, outer contour, inner contour, scanning mode, scanning speed, beam correction, shrinkage compensation, scanning line sequence, filling spacing, filling direction, laser power, (off) focus, surface filling Parameters, offsets, etc. It can be seen that it is very difficult to control and balance more than 160 variables that affect processing quality through human experience.

AI releases industrialization potential

Fortunately, artificial intelligence (AI) has made great progress. Let a machine take over some quality assurance tasks in AM. This sounds a bit outrageous. According to market observations from 3D Science Valley, it turns out that in AM additive manufacturing The problems with the monitoring data collected during work are very different from the data collected by offline tests (such as CT scans or ultrasound tests). The offline test data characterizes the characteristics of the final AM additive manufacturing component, while the monitoring data only characterizes the characteristics of the specific layer of construction.

In additive manufacturing, there is still a great need to reduce the cost of the parts produced. This is related to a process that may consist of hundreds of thousands of layers of processing. According to the market understanding of 3D Science Valley, offline CT testing will not only increase the overall cost, but also limit the geometry, because the parts must have the appropriate shape to be scanned and tested. If offline monitoring is replaced by intelligent in-process monitoring and testing, this opens up new space and may reduce overall costs.

According to the market understanding of 3D Science Valley, a process monitoring system with integrated AI will support this transformation and realize a direct method that evolves from full-detail testing to intelligent testing. International companies that use AI to control the quality of additive manufacturing include printsyst in Israel, addiguru in the United States, nebumind in Germany, and Nnaisense in Switzerland.

Better quality

In cooperation with Nnaisense, EOS developed an AI to simulate the EOS LPBF process as the first step to achieve automatic quality monitoring. Allows prediction of hierarchical monitoring data, which is collected by the process monitoring system. Comparing forecasts with actual results allows for faster assessment of violations in the process. This helps to better understand what to expect in terms of quality data before building work. Users can also understand how different part designs and process settings affect sensor readings, and as artificial intelligence continues to learn from the process, this will improve accuracy.

© EOS

Currently, inspired by PyraMiD-LSTM (Parallel Multidimensional Long Short-Term Memory), convolution and loop processing are used. Powerful parallelization enables EOS and Nnaisense to calculate about 1000 layers of predictions in less than 2 hours through the GPU with a resolution of 4MP per layer.

The original monitoring data is compared with the predicted data, and then the impact of the predicted results on the processing results can be considered in more detail in the quality assessment. Of course, AI must be trained for different material categories. According to the understanding of 3D Science Valley, different materials have different processing requirements and sensitivity to the process. AI can not only learn the effects of overheating of different materials in some known overhangs and narrow positions, but also learn the effects of air flow distribution in the machine and the placement of parts on the processing results. Another use case of the AI ​​developed is to optimize scanning strategies or process settings to receive more uniform monitoring results, thereby obtaining more uniform part quality. Moreover, these do not need to build a single job, use precious machine printing time and waste precious resources (powder, energy, testing, manual labor…). Achieving this goal will truly change how to optimize AM additive manufacturing production operations today and take an important step towards responsible manufacturing.

Huge room for innovation

According to the judgment of China’s Oz Asia Pacific, with the support of technology such as simulation and digital twins, additive manufacturing will break through the bottleneck, give full play to its advantages, and realize the huge space for innovation that people expect. But with the development of additive manufacturing technology, additive process simulation technology will also continue to improve. The future development trends are mainly in the following directions:

• Macro-scale additive process simulation will become more popular and applied in engineering. Additive process simulation will be gradually introduced in the full cycle of additive design, process and manufacturing to ensure the printability of designed products;
• Material-equipment-printed parts-support design and process design-process parameter package-macro characteristics-micro characteristics-post-processing-performance prediction, the entire process will be streamlined and platformized;
• Mesoscopic analysis and microanalysis will gradually move from the research and scientific research stage to the engineering application;
• Support design and optimization software driven by physical process simulation will be gradually available;
• Using test data and simulation data, AI algorithms and multi-scale algorithms will realize offline prediction of additive processes;
• More metal material data will be tested and entered, and more metal additive process methods will be simulated.

(Editor in charge: admin)