Advantech WISE-PaaS/AFS: A Complete Path for AI Multi-Model Training, Inference, and Updates
Speed up Development of AIoT Solutions with an Edge-to-Cloud-Ready Unified AI Platform
WISE-PaaS/AFS Helps to speed up the AI applications: Going from 0 to 1 in POC smoothly, Quickly expanding from 1 to 100
Jamie Su, product manager of Advantech IoT.SENSE, emphasizes that the company hopes to help users speed up the AI POC process with WISE-PaaS/AFS; going from 0 to 1 smoothly, quickly expanding from 1 to 100, and letting AI applications come to life.
Generally speaking, the process of building an Industrial AIoT solution includes model training (cloud) and inference (edge). For the latter, Advantech has accumulated more than 30 years of Edge Intelligence experience so that WISE-PaaS/AFS stably supports both inference engine operations and remote model deployment. Due to the large number of machines utilized in modern industrial scenarios, it is necessary to adopt "one machine, one model" predictive inference. Using WISE-PaaS/AFS's OTA remote deployment function reduces the burden on deployment personnel and makes it easier to promote the retraining and optimization of existing models.
Convenient for SI/DFSI (Domain-focused Solution Integrator) integration, WISE-PaaS/AFS can build the inference engine as a RESTful API and it supports Windows, Linux, and Android operating platforms to quickly interface with heterogeneous systems and customize the best AIoT application scenarios for customers. The AFS inference engine can be packaged as a Docker Image or EdgeX Foundry—both of which support remote deployment.
Since AFS runs on the WISE-PaaS industrial IoT platform, it inherits WISE-PaaS and supports the characteristics of public, private, and hybrid clouds. For public clouds, it supports Azure, AWS, and Alibaba Cloud. For private clouds, it is based on Advantech WISE-STACK software and a hardware-integrated edge intelligence private cloud solution. For hybrid clouds, public clouds paired with the Advantech GPU server allows GPU computing to run on its own platform, thereby reducing computing costs.
Jamie Su emphasized that WISE-PaaS/AFS services contain a number of important functional modules. First, the multi-model training module is based on Hadoop Yarn and Kubernetes technologies to help users realize flexible dividing, scheduling, and management of computing resources. Secondly, AFS supports multiple databases such as PostgreSQL, InfluxDB, MongoDB, and Ceph. It can also interface with WISE-PaaS/APM data sources, enabling data scientists and AI engineers to quickly integrate various types of data for model building and training.
The Workspace module integrates the Jupyter Notebook framework in advance, making it easy for users to develop algorithms directly online. Secondly, it provides an offline development environment. Even if users write code in non-Python language, the code can still be packaged as a Docker Image and uploaded to AFS for online execution. Other modules such as Catalog, Task, Model Board, and Inference can help users to subscribe to third-party open source algorithms, set up automatic scheduling of model retraining and deployment, view model training performance through visual tools, and manage large numbers of models and multiple sources of information respectively.
WISE-PaaS/AFS Framework Supports both Cloud and Edge Inference
Dr. Ali Chang, assistant project manager and data scientist for Advantech IoT.SENSE, said that Advantech hopes to gain support from many ecosystem partners and DFSI through the launch of the WISE-PaaS/AFS platform. Advantech has set out to provide three AI solutions, hoping that these AI solutions will lead partners to develop more AI applications. The first solution is AOI (Automated Optical Inspection), which uses deep learning technology to automatically identify the types of defects and reduce the cost of manual inspection.
The second solution is PHM (Prognostic and Health Management), which provides machine prognosis services for specific production equipment. For example, it measures the production signals of machines. To achieve the goal of near-zero failure, when performance is gradually declining, appropriate maintenance steps are taken immediately. The PQA (Predictive Quality Analytics) solution analyzes production process parameters and key quality indicators to establish production quality prediction models and optimize manufacturing processes. There is high demand for this feature across a range of industries: textiles, semiconductor factories, petrochemicals/cement, plastics, and machinery.
Dr. Michael Chuang, assistant project manager and data scientist at Advantech IoT.SENSE, reiterated that WISE-PaaS/AFS is a comprehensive solution, from edge to cloud, with five key features. The features include support for various open source libraries for AI model training during the development phase, making it easy for users to perform online development. Because models need to be continuously optimized in the verification phase, it also supports multi-model management tools; meeting the dual requirements of scheduling retraining and model management. WISE-PaaS/AFS supports cloud inference in the online phase (especially for the vast field of smart city applications). It integrates a large number of edge computing devices and performs model delivery to achieve edge inference in the deployment phase.
WISE-PaaS/AFS has a well-documented track record of delivering value for its partners and customers. In fact, Advantech's own Linkou PCB factory uses the AI AOI solution for defect detection. After half a year of operation, the detection rate is nearly 100% and the production line’s detection efficiency has increased by more than 33%. Advantech also developed a passive component defect detection solution with a partner customer and sold it through the WISE-PaaS Marketplace. The end user can now automatically deploy the relevant model to its edge MIC-730 AI system by subscribing to the solution, which will begin to perform inference and detect defects in each passive component.
Developers across many fields—such as smart factories and cities—can make use of the WISE-PaaS/AFS framework to implement AI projects, upload data from edge to cloud, build inference engines in the cloud, and automatically deploy to the edge computing platform. Each link in the process has complete product support. Developers no longer need to open up the entire link from edge to cloud and the link from the cloud to edge devices. WISE-PaaS/AFS offers an optimal shortcut for enterprises moving towards AIoT and comprehensive digital transformation.