The integration of new-generation information technologies including artificial intelligence, big data and the Internet of Things with traditional manufacturing technologies has not only boosted manufacturing development, but also made intelligent manufacturing a core development target for traditional manufacturers. Nowadays, unblocking the core industrial links is the key for enterprises to truly realize intelligent manufacturing.

A new manufacturing revolution driven by intelligent manufacturing, Industry 4.0, Industrial Internet of Things (IIoT) and other concepts is in full swing. In 2019, mobile communication technology officially entered the commercial deployment phase. Leveraging the technical characteristics of 5G, alongside other communication technologies and innovative components such as Augmented Reality (AR), Artificial Intelligence (AI), machine vision, deep learning and cloud services, enterprises are expected to rapidly connect the intelligent front-end workshop operations with back-end cloud management systems, helping them fulfill the Industry 4.0 vision of intelligent manufacturing.

Conception 1: The "Ren Vessel" — AI Penetrates End Devices

One of the ultimate goals of intelligent manufacturing is to build self-managing unmanned factories. Therefore, embedding artificial intelligence, machine learning and other capabilities into on-site production equipment to enable autonomous learning and self-troubleshooting has become a major development direction for intelligent manufacturing. Nevertheless, building AI and machine learning systems is no easy task, requiring specialized talent to tailor machine learning environments for diverse production sites.

Huang Zhenwei, Solutions Architect at Amazon Web Services (AWS), stated that the whole machine learning workflow involves unavoidable steps: algorithm selection, environment construction, data input for model training, and simulation verification of training results against expected outcomes. Even with dedicated data and machine learning specialists, the full training cycle still takes 3 to 6 months. For enterprises with no relevant in-house expertise, the timeline will be far longer. Fortunately, Amazon opens all its machine learning and AI services to clients and continuously upgrades its toolset to help businesses quickly build proprietary machine learning systems.

Notably, high-quality data is essential for training reliable machine learning models. Huang pointed out that collecting valuable information is critical to improving model accuracy. Enterprises must first identify what data to gather to generate usable training datasets, and this screening process consumes substantial time. Poor data preparation may lead to failed or underperforming models in later stages. To address this pain point, Amazon provides end-to-end services to support enterprises through every stage of model development, facilitating the delivery of usable, precise and self-learning machine learning systems.

Intelligent Upgrade of End Devices

Intelligent manufacturing equips factory machinery with condition sensing, autonomous management and decision-making capabilities, with the ultimate goal of boosting overall production efficiency. This means both on-site equipment and central control platforms require intelligent transformation. Machine vision serves as a core driver of productivity improvement; combining machine vision with artificial intelligence has become an irreversible trend to deliver clearer image capture and more accurate defect judgment for industrial equipment.

Zhou Kunren, General Manager of G4 Technology, explained that machine vision was invented to replace human visual inspection. It cuts labor costs and expenses incurred by defective product recalls while raising production throughput. Massive inspection data accumulated via visual detection can be combined with statistical models to guide corporate quality management and strategic decision-making, making machine vision increasingly vital and widely deployed on production lines.

The four mainstream applications of machine vision include Guidance (alignment recognition), Inspection (defect detection), Gauging (dimension measurement) and Identification (text & barcode reading). However, existing machine vision technology is limited by low resolution for complex images and incapability of depth and thickness measurement. Deep learning AI algorithms paired with 3D measurement technology can break through these bottlenecks. According to Zhou, the integration of machine vision, AI and 3D measurement enables equipment to identify subtle defects in complicated images and integrate thickness & depth inspection functions on a single machine, a critical component of fully automated Industry 4.0 factories. The era of AI-powered 3D measurement in industry is fast approaching.

Statistics from Research and Markets show that AI and 3D measurement technologies will outpace the overall growth of traditional machine vision. From 2017 to 2022, the global machine vision market posted a Compound Annual Growth Rate (CAGR) of 8.15%, AI vision software hit a CAGR of 49%, and 3D vision products achieved an 11.07% CAGR. The firm forecasts the total machine vision market to reach USD 144.3 billion by 2022, with AI vision software valued at USD 997 million and 3D machine vision products at USD 2.13 billion.

Beyond machine vision, AI is being embedded into all types of production equipment. Yu Wenhong, Asia Pacific Technical Marketing Manager at STMicroelectronics, noted that motor training, sensor fusion and voice control all fall under AI applications. Most of these use cases still rely on manual model training and can run lightweight AI algorithms on Microcontroller Units (MCUs).

In fact, running AI algorithms to empower intelligent end devices is only a recent new function of MCUs. As distributed "mini-brains" across the intelligent manufacturing architecture, MCUs govern motors, Human-Machine Interfaces (HMIs), communication modules, cloud connectivity and digital power control. Yu added that factories have long depended on MCUs to handle operational tasks; the rollout of intelligent manufacturing has driven surging MCU adoption, as these chips can execute simplified AI algorithms and prompt manufacturers to shift from Microprocessor Units (MPUs) to MCUs.

Daniel Ho, Distribution Sales Manager at Analog Devices (ADI), highlighted that all digital data originates from analog signals. For example, sensors convert analog current and voltage readings from machinery into digital data to bridge the physical analog world and the digital machine environment. Low-precision converters generate useless "garbage data" that severely impairs AI and machine learning performance.

To avoid wasted investment and subpar AI outcomes, enterprises upgrading factory equipment intelligence must look beyond MCU AI execution capabilities and refine core peripheral components such as signal converters.

Moreover, comprehensive collection and analysis of operational data from edge devices relies on a wide range of hardware components. Yan Jinfu, Senior Applications Engineer at Maxim Integrated, emphasized that sensors, I/O modules, transmission hardware and power components are indispensable for building secure, complete intelligent manufacturing systems.

Most industrial equipment operates on 24V power, while many new systems are migrating to 48V. Both voltage standards face surge risks when miniaturized, high-density intelligent systems are required. Designers must integrate external voltage divider resistors and isolation components to prevent equipment damage and operator injuries caused by electrical surges.

Conception 2: The "Du Vessel" — From On-Site Edge Nodes to Cloud Platforms

After realizing intelligence for front-end factory equipment, the next challenge lies in aggregating data from discrete production machines, transmitting it to the cloud for analysis, and feeding analytical insights back to equipment for iterative learning. Factories currently deploy hundreds of incompatible transmission technologies, creating costly compatibility and deployment hurdles for manufacturers.

Lin Zhongheng, Senior Applications Engineer & Technical Committee Member at Texas Instruments (TI), stated that wireless technology selection varies drastically by application scenario. Sub-1GHz solutions suit long-distance, battery-powered industrial devices requiring high robustness; Bluetooth delivers high-speed high-volume data transmission and mobile control; Thread, Wi-Fi and other protocols each carry unique strengths and applicable use cases. Equipping separate MCUs for multiple wireless standards leads to overly complex system design and inflated costs.

To resolve this pain point, MCU vendors now support multi-standard and multi-protocol chips to simplify product development for manufacturers. Chiang Chih-liang, Business Manager at Silicon Labs, observed that IoT products for both industrial and smart home markets increasingly demand multi-protocol compatibility, and single-chip multi-protocol integration can cut system costs by an estimated 40%.

System-in-Package (SiP) Chips Enable Multi-Protocol Integration

System-in-Package (SiP) technology offers a practical path to build multi-protocol chips for intelligent manufacturing and IIoT use cases. Cheng Min-yao, Director at ASE Group, explained that unlike System-on-Chip (SoC), SiP integrates cutting-edge sub-system components including memory and RF modules, plus specialized parts such as oscillators, without process compatibility constraints faced by monolithic SoC designs.

SiP technology also delivers compact, low-cost system chips that support flexible component combinations for diverse IoT applications. Nevertheless, Cheng acknowledged that SiP has limited flexibility for low-volume, highly diversified IoT chip demands, and is best suited for standardized integrated modules such as combined RF and MCU units.

Simulation Tools Simplify Complex 5G Design

Short-range wireless mesh networks cover on-site factory communication, while 5G serves as the critical link connecting manufacturing plants to cloud platforms. Wei Peisen, Regional Technical Manager at Ansys, outlined transformative 5G features: ultra-low latency, ultra-high throughput, expanded coverage and wide bandwidth. These capabilities enable AR, Virtual Reality (VR), AI, machine learning and other innovative applications in intelligent manufacturing while establishing stable factory-cloud connections.

However, 5G design introduces unprecedented technical complexity, including Massive MIMO, novel modulation schemes, phased antenna arrays and unfamiliar millimeter wave bands, posing major challenges for engineers working on antennas, bare dies, packaging and PCB layouts.

Early-stage simulation and product verification allow engineers to optimize designs and avoid costly late-stage redesigns triggered by discovered defects. Wei stressed that despite the upfront cost of simulation software, the savings from eliminated rework far outweigh the investment. Simulation platforms enable designers to validate structural and material parameters before physical hardware prototyping, eliminating unsatisfactory finished products. Modern simulation tools incorporate big data analytics and cloud computing, delivering accuracy rates above 90% and accelerated computation speeds to resolve core 5G design challenges and boost engineers’ advanced design capabilities.

Accessible Cloud Platform Architecture

Most enterprises recognize the need for cloud systems to support data analytics and operational management when deploying IoT and Industry 4.0, yet many lack specialized teams or capital to build proprietary cloud infrastructure. Companies with deep expertise in networking and e-commerce have launched mature cloud construction solutions to reduce implementation burdens for manufacturers.

Lyu Xinyu, Senior Business Development Manager at AWS, noted that IIoT and cloud integration delivers tangible industrial benefits: mistake-proof production, reduced rework, fewer on-site supervisors, improved product quality and customer-centric customized manufacturing. Developing a complete end-to-cloud industrial system that balances practicality and usability remains a complex task, but AWS leverages Amazon’s ecosystem to provide comprehensive hardware and software tools covering edge devices, gateways and cloud services. Its engineering team offers full-cycle technical support to align solutions with client requirements and address core concerns such as secure data transmission.

Seamless Integration of Legacy and New Systems

A common concern among enterprises building smart factories is whether existing production equipment and management systems must be fully replaced, and how to unify oversight of legacy and newly purchased assets. Pan Huangliang, Associate Director of Software & Solution Product Division at Axiomtek, pointed out that Industrial Personal Computers (IPCs) differ from consumer PCs through robust environmental resistance and wide operating temperature ranges, granting them far longer service lifespans.

Upgrading software instead of fully replacing hardware represents a cost-effective strategy to connect old and new factory systems without massive architectural overhauls. As an example, retrofitting legacy machines with sensors and wireless communication modules, paired with new software on existing IPCs, enables enterprises to collect and analyze sensor data and centrally manage all connected equipment with minimal new hardware procurement, accelerating the factory’s intelligent transformation.