Smasoft’s SmaAI & Advantech's WISE-PaaS/AFS: Realizing Mass Deployment for AI Applications
Smasoft Technology focuses on the development of automation software solutions using the Advantech WISE-PaaS/AFS (AI Framework Service) as a scale-up platform to help SmaAI’s deep learning defect inspection tools automate models for large-scale deployment to production lines. The AI defect inspection solution, powered by SmaAI and WISE-PaaS, significantly enhances overall manufacturing productivity and efficiency.
Automated Training and Large-Scale Deployment: The Keys to Applying AI to the Production Line
AI has become an increasingly powerful tool for manufacturers who need to meet demand for small-volume and large-variety production, improve manufacturing quality, and pass on process experience. Many manufacturers rely on imported AI. While the results of AI proof of concept (POC) in these instances are often acceptable, the process of moving towards the use of AI in production lines at scale frequently fails.
Jesse Chen, the General Manager of Smasoft Technology, points out that many companies send personnel to specialized schools to train in AI modeling analysis. While they might be able to make accurate models in the POC stage, the models can rarely be applied to the actual production stage due to a limited understanding of broader factory ecology. Implementing AI smoothly and at scale proves to be too great of a challenge. Even if development advances beyond the POC stage, further challenges remain: it's extremely time-consuming and laborious to sort through images, implement training, and copy models prior to deployment. Most companies are limited by the manpower of their engineers—hindering the process of deploying AI solutions at scale.
In order to scale up AI manufacturing, Smasoft decided to cooperate with Advantech to integrate its SmaAI deep learning defect inspection tools into Advantech’s WISE-PaaS/AFS (AI Framework Service) to create a unique niche in automated training and deployment that was capable of breaking down traditional scaling barriers. Through this partnership, Smasoft created an automated inspection solution that was well received by manufacturers across an array of tech industries.
Jesse Chen says that Smasoft focuses on the development of automation software solutions designed to help customers reduce development time and improve work and production efficiency. Releasing the SmaSEQ workflow automation tool is one example of how Smasoft lets customers easily configure application processes such as motion control, vision, and I/O without having to write complex and time-consuming underlying programs. SmaAI is based on a similar concept; customers who do not understand AI still have the opportunity to implement defect detection through the GUI and model provided.
Smasoft's SmaAITrainer GUI supports models created in-house as well as ones provided by third parties. It performs model training through the Advantech MIC-770 computer and a GPU card. In the initial POC stage, AI engineers collect pictures, classify them, execute training programs, and verify training results. If the results meet expectations, then the model is manually copied and loaded in the computer (for example, the Advantech MIC-730AI) on the production line to perform inference. If the process goes smoothly in the POC stage and the model needs to be deployed at scale to production line machines, it will cost a lot of manpower, time, and money. It is also difficult to manage.
Connecting Edge AI to WISE-PaaS/AFS Forms a Complete Device-to-Cloud Solution
Smasoft's original intention was to develop functions that allow the AI model to be executed in the cloud to meet the needs of training, deployment, and updates of production line automation. After finding Advantech, both parties decided to cooperate by putting the SmaAI engine into the WISE-PaaS/AFS. After the company classified its images, training programs could be uploaded to WISE-PaaS/AFS to continuously search for images, train, and update the model on the back-end. Deployment to the production line was performed automatically to greatly reduce the level of manpower required for model training. WISE-PaaS/AFS helped customers overcome many of the obstacles from POC to large-scale deployment.
Companies can now connect to the WISE-PaaS/AFS cloud platform via Advantech Edge AI (including MIC-770, MIC-730AI), then integrate with Smasoft's SmaAI for complete device-to-cloud model training and deployment verification solution; allowing companies to accelerate their AI deployment without the need for additional components.
WISE-PaaS/AFS functions are continuously updated and improved. Advantech recently released the Hyperparameter Tuning function. If a company wishes to deploy a model to hundreds of devices at the same time but only relies on experienced engineers for the time-consuming task of tuning parameters, efficiency will suffer. Companies retraining models can sometimes struggle to improve their accuracy—wasting valuable time on repeated parameter tuning. The WISE-PaaS/AFS hyperparameter adjustment function helps the AI algorithm automatically search for optimal parameters to greatly shorten the manual trial and error process. Tuning algorithms currently supported include TPE, Random Search, Anneal, Naïve Evolution, SMAC, Batch Tuner, and Grid Search.
Smasoft SmaAI x Advantech WISE-PaaS: An Optimal AI Defect Detection Solution
The AI defect detection solution provided by Smasoft is a typical industrial app (I.APP) as defined by Advantech. Smasoft's considerable know-how in factory automation technology allows for deep discussions with customers about light source, cameras, motors, and robot arm control methods. Smasoft is an experienced Domain-Focused Solution Integrator (DFSI) with formidable domain knowledge in both the IT and OT markets. Looking to the future, Smasoft hopes to continue working closely with Advantech in creating high-quality AI solutions that resolve pain points and achieve the benefits of AI application at mass scale.
WISE-PaaS Industrial IoT Cloud Platform
Edge-to-Cloud AIoT Architecture to Enable Real AIoT-Powered Cloud Business Models.
IoT Software Online Business Channel to Buy, Deploy, and Customize Industrial Software Needs.