Improving AI Model Accuracy with Synthetic Data
The manufacturing industry continues to embrace new and ground-breaking technologies, one being synthetic data for training industrial machine vision systems. Computer vision powered by AI is helping to solve many industrial use cases, such as package inspection, defect detection, predictive maintenance and much more. There is a huge opportunity among manufacturers to reap the benefits that AI (Artificial Intelligence) and ML (Machine Learning) have to offer, including improved profitability, efficiency, and safety.
SynthAI is a product that allows anyone to upload CAD files of parts or objects which are translated into synthetic data. The cloud-based platform can also train various types of machine learning models that are able to detect these parts.Zachi Mann, R&D Manager for Tecnomatix at Siemens Digital Industries Software
In a recent podcast featuring Zachi Mann, robotics expert for SynthAI, a cloud solution that easily generates synthetic data to train vision machine learning algorithms from CAD models for industrial vision systems, we discuss how although machines are becoming intelligent enough to make their own decisions, data collection and annotation remains a complex issue and bottleneck. That is because data must be collected and cleaned, and simulations must be built and validated, resulting in a labor intensive and costly exercise. Instead, engineers can use synthetic data as an inexpensive alternative to real-world data that is increasingly used to create accurate AI models.
Synthetic data bridges the gap between Robotics and Simulation
Industry use cases for this type of technology include kitting, sorting, picking, shop floor safety, throughput analysis, quality inspection, and many more. Robots, for example, are involved in performing complex tasks such as automating processes, detecting objects, picking, and placing. These tasks are powered using technology like computer vision algorithms and other types of sensors. These processes also possess enormous amounts of data which can be collected and utilized to increase efficiency, optimize throughput, and improve production quality. However, data collection is a timely and cost-prohibitive task. There is also a question of scale, meaning how much data is sufficient to create and train an accurate AI model to perform a specific task. This often leads to manufacturers choosing to invest more time in collecting data and training the AI or partnering with service companies to create realistic simulations.
Robotics expert, Zachi Mann, explains how synthetic data can be a massive solution to reduce the time, human capital and money involved in commissioning ML systems with real world data. Discover the cloud-based solution, SynthAI, that can help manufacturers quickly deploy AI and ML techniques to their processes and reduce the time to market with increased model accuracy.
Synchronizing Synthetic Data with the Digital Twin
Synthetic data can also be used for building accelerating and validating processes within a simulation environment. A digital twin-enabled robotic station which can detect parts to be virtually simulated by feeding synthetic images of the part’s CAD (Computer Aided Design) model generated using the SynthAI platform.
With an ML model trained to perform a specific task the completeness check of the process can be carried out. Additionally, the throughput of the process is also determined and validated before going into production. The synthetic data generated from Siemens SynthAI in conjunction with existing manufacturing systems in Siemens portfolio like NX, Tecnomatix and simulation systems like Process Simulate allows engineers to create scenarios, simulate and validate processes, combine synthetic data with real data and create and train accurate AI models before executing them in the physical environment.
Listen to the AI Spectrum podcast featuring robotics expert, Zachi Mann to learn more about SynthAI, a new synthetic data generation solution from Siemens.