Issue link: https://iconnect007.uberflip.com/i/1207026
22 SMT007 MAGAZINE I FEBRUARY 2020 For example, some tools allow manufac- turers to simultaneously deploy programs and inspection conditions across multiple lines, which enhances productivity and, more importantly, data integrity with consistent performance. Operators can further improve line maintenance with other tools for real-time monitoring to instantly display relevant pro- cess parameters at remote locations for imme- diate analysis and action. What's more, com- bining multipoint views from SPI, pre-reflow AOI, and post-reflow AOI with real data man- agement and monitoring allows operators to determine actionable insights to optimize the processes. However, the adaptation of AI- powered process tools takes optimization to a higher level. Converting all the data requires a simulation tool to review identified defects with accumu- lated historical data from printed circuit board assembly (PCBA) lines while avoiding unnec- essary downtime. Software tools can reliably allow manufacturers to predict the effects of fine-tuning without stopping the line. Moving forward, an AI-powered platform can autono- mously render complex process optimization decisions typically reserved for dedicated pro- cess engineers. Embracing connectivity can create a smart factory. For instance, software modules can exercise complex algorithms to develop closed-loop process recommendations. The machine-to-machine (M2M) connectivity drives the smart factory vision one step further by enabling automatic SMT line maintenance. Finally, combining inspection with printers and mounters can enable the network tools to con- nect and simplify communication across the entire PCBA line. Defining the correct process parameters often requires a high degree of expertise because of the various environmental considerations affecting the process. Using AI-powered sys- tems and M2M connectivity, manufacturers can link inline inspection systems with the associ- ated printer and mounters in the line to over- come the challenges. Figure 2 shows how auto- mated machine learning can already match the results from process experts, and this will only improve. Although 2D AOI is still a technology in the market, more manufacturers are adopting 3D AOI to increase board quality. The benefits are clear. Using clearly defined thresholds backed by accurate data will eliminate the need to constantly debug inspection programs. More- over, measurement data generated from some 3D AOIs provides meaningful insights about the process and helps eliminate the root causes of a defect. Combining a 3D SPI with 3D AOIs enables manufacturers to accurately control and monitor the board assembly process. But with so much data, engineers are hard- pressed to collect, process, and implement all the data using traditional techniques and soft- ware. AI and deep learning lay the foundation for machines to learn from the vast amounts of process data collected by adjusting the out- put based on the data inputs and performing tasks to help engineers perform tasks more intelligently. The many examples we hear about, such as computers playing chess or autonomous (self-driving) vehicles, use deep learning to achieve tasks by processing large amounts of data and recognizing patterns in that data. This is ideal for volume PCB production and helps create a data set for a smart factory. From statistical process control to instant pro- gram refinements, AI-powered platforms can intelligently apply real-time data to improve production processes. Going beyond smart factory solutions, manufacturers can use the same technology to optimize the process and adjust process parameters by exercising com- plex machine-learning algorithms. Realizing a smart factory means taking a practical approach to processes and systems while examining areas to improve productiv- ity. Combining machine learning with 3D mea- surement data generated during inspection helps manufacturers define inefficiencies and boost line efficiency. Machine learning uses programmed algorithms that receive and ana- lyze input data to predict output values within an acceptable range. As new data is fed to these algorithms, they learn and optimize their operations to improve performance, develop- ing intelligence over time.