"The intelligent upgrade of the industry is driving a comprehensive reconstruction of the underlying infrastructure, and the infrastructure upgrade for industrial communication must precede AI deployment. In other words, without network upgrades, everything else is merely empty talk." In an exclusive interview with China Control Engineering, Zhang Heng, Product Director of Moxa China, pinpointed a widely overlooked issue in the current intelligent transformation of the manufacturing sector.
Today, AI (Artificial Intelligence) is accelerating its descent from the cloud to the production site, driving industrial upgrades from the past superposition of single-point applications to a full-chain upgrade encompassing perception, analysis, decision-making, and execution. As an expert in industrial network communication, Moxa has always been at the forefront of technology implementation, sensing every iteration of market demands. In this interview, Zhang Heng shared observations and reflections from the front lines of practice, focusing on core topics such as the changing demands for industrial networks in the AI era, scenario implementation, and security governance.
From Connectivity Pipelines to Intelligent Foundations: AI Reshapes the Positioning of Industrial Networks
In the past, the primary task of industrial networks was to achieve data connectivity and single-point control. From the early days of completing device connections based on industrial switches, to the continuous promotion of IT and OT integration and cloud-edge collaboration, and then to the collaborative hosting of AI, robots, edge computing, and data, and finally moving towards the network foundation for AI factories, the development trajectory of demands in China's industrial communication market is clearly visible.
As the manufacturing industry enters a new stage of accelerated AI implementation, manufacturing customers are no longer satisfied with mere device connectivity. Instead, they require the network to support the efficient collaboration of multiple types of business, enabling various complex and precise control data to be provided with a stable, synchronized, and deterministic operating environment through an "end-to-end network" industrial network. "Industrial networks are no longer just communication devices in factories; they have become the fundamental network foundation for industrial AI," Zhang Heng stated.
The fundamental driving force behind the changing demands lies in the fact that AI is reconstructing the operational logic of production systems. When perception, analysis, decision-making, and execution form a complete intelligent closed loop, AI transitions from a tool for auxiliary analysis to a core component of the production system, and the role of the network is upgraded accordingly. Zhang Heng summarized the irreplaceability of industrial networks in the AI production closed loop into four core dimensions:
01. First, turning on-site perception into usable real-time data. Industrial networks connect sensors, PLCs, robots, AI vision, and equipment, ensuring that on-site data is delivered to AI analysis systems in a stable and low-latency manner;
02. Second, bringing AI decisions back to the production site. The judgments generated by AI ultimately need to be translated into equipment actions, such as adjusting robot paths, changing process parameters, or triggering equipment control. The industrial network serves as the critical bridge between AI and OT control systems;
03. Third, ensuring real-time performance and determinism. No matter how accurate AI decisions are, if there is communication delay, packet loss, or network interruption, the decisions cannot be reliably executed;
04. Fourth, keeping the entire network continuously visible, manageable, and protected. Against the backdrop of the high integration of AI, OT, and IT, the network must provide traffic visibility, equipment status monitoring, fault diagnosis, and cybersecurity protection to ensure the long-term stable operation of equipment.
This is precisely the aspect currently widely overlooked by most manufacturing enterprises in AI implementation. Without a solid underlying network foundation, AI decision-making is out of the question. Network latency, data packet loss, and time asynchronization among equipment—every tiny error can lead to disastrous consequences in production decision-making.
"The value of industrial networks lies not only in connecting AI and equipment but also in ensuring that AI judgments can be reliably delivered with the correct data at the right time, forming a sustainable closed-loop process of perception, analysis, decision-making, and execution," Zhang Heng summarized. "In other words, AI decides what to do, while the industrial network determines whether it can be done reliably in real time."
From Reliable Connectivity to Deterministic Communication: Technological Upgrades in Industrial Networks
As the core foundation for AI implementation scenarios, the capability boundaries of industrial networks affect whether AI decisions can be reliably executed. This also drives industrial networks towards predictable and controllable deterministic transmission, making the prevention of latency jitter, support for the coexistence of multiple types of traffic, and guarantee of reliability in mobile scenarios imperative issues to be solved. This change directly refreshes the capability benchmark for OT networks. Traditional production lines require reliable control, with the core being stable connections, whereas after AI enters the production system, the core requirement is upgraded to predictable data and control closed loops.
Upgraded Requirements for Latency and Jitter
In high-precision scenarios such as AI vision servo control, AI models can complete hundreds of inferences per second. A latency jitter of just a few milliseconds can lead to trajectory calculation deviations, ultimately resulting in product scrapping or even equipment collisions. Therefore, the importance of low latency, low jitter, and time synchronization has been elevated to an unprecedented height in the AI era.
Deterministic Transmission of Multiple Types of Traffic
AI vision generates massive amounts of image data, and inference models require large-scale data exchange. Meanwhile, critical control data such as PLC and motion control are also transmitted on the same network, and various types of traffic must not interfere with each other. This requires the network to possess multiple capabilities such as traffic isolation, QoS (Quality of Service), TSN (Time-Sensitive Networking), redundancy, and priority management to ensure the deterministic transmission of critical business.
For traditional production lines, network stability is sufficient, but for applications in AI scenarios, every millisecond of network latency must be predictable and controllable. This is also the most profound change that AI has brought to industrial networks.
On the wireless side, as applications such as generative AI, digital twins, AI vision inspection, and autonomous mobile robots rapidly penetrate the manufacturing site, factories are evolving into AI-driven physical intelligence. Wireless technologies such as industrial Wi-Fi and industrial 5G have taken center stage with their flexible deployment and ability to adapt to the needs of flexible manufacturing and dynamic expansion. They are currently being applied at scale in key scenarios such as AGVs (Automated Guided Vehicles), overhead cranes, and industrial robots.
However, the particularity of wireless networks also brings new challenges. Wireless networks are invisible and intangible, easily affected by various environmental factors, making anomalies difficult to detect in time, often only noticed when operations are hindered. At the same time, communication interruptions when equipment switches across AP (Access Point) zones in mobile scenarios can directly cause robot shutdowns and inspection frame drops, thereby affecting production capacity and yield rates. Therefore, roaming quality in mobile scenarios has become the most critical pain point for enterprises deploying industrial wireless networks today.
Addressing these series of challenges, Zhang Heng introduced two core capabilities of Moxa in wireless technology. The first is Turbo Roaming technology, which assists mobile devices such as AMRs (Autonomous Mobile Robots) in maintaining connection stability when switching between different APs with high-speed roaming switching capabilities of less than 50 milliseconds. The second is AeroMesh technology, featuring a multi-path redundancy architecture that enables self-organizing networks and automatic repair. When a communication path or network node experiences an anomaly, the system can find an available path and restore communication within 2.5 seconds.
Currently, these two technologies have been implemented in multiple AI scenarios. For example, in the scenario of autonomous mobile robots equipped with AI vision, the robots need to continuously transmit high-definition images, sensor data, and control instructions when moving across workshops. After introducing Turbo Roaming, it effectively meets the needs for stable transmission of visual data and real-time control of autonomous mobile devices. In the intelligent inspection scenario of high-tech factories, Turbo Roaming technology enables high-speed roaming switching of less than 50 milliseconds, supporting the real-time and stable transmission of images and sensing data during long-term autonomous inspections.
During the interview, Zhang Heng specifically mentioned that industrial wireless cannot only focus on transmission performance but must also promote the shift of operation and maintenance (O&M) models from post-fault processing to proactive O&M. Relying on Moxa's MXview One global visualization management platform, enterprises can master the status of network-wide equipment, security conditions, and anomaly alerts in real time, moving troubleshooting forward to prevent faults from expanding and affecting production. The supplementation of this proactive O&M capability also ensures that the value of industrial networks is no longer limited to communication connections but extends to lifecycle stability assurance.
From Security Foundation to Global Expansion Empowerment: Cybersecurity Advances Towards Full-Lifecycle Governance
In the AI era, the number of connected devices continues to grow, and the degree of software dependence is deepening, expanding the attack surface of industrial networks accordingly. Meanwhile, the EU's Cyber Resilience Act (CRA) is being progressively advanced, and market requirements for equipment suppliers have extended from product security to continuous governance capabilities throughout the entire product lifecycle. These dual pressures are profoundly changing manufacturing enterprises' cognition and selection criteria for OT security.
"OT security is not about purchasing the most secure equipment available today, but about choosing a supplier that can continuously discover, respond to, patch, and prove its security throughout the equipment's lifecycle of over ten years. This is a long-term commitment from the equipment supplier and also a very core and critical capability," Zhang Heng revealed the essence of OT security.
Security is not a one-time product delivery but a systematic capability over a long cycle. Therefore, Moxa not only provides industrial security-related products such as firewalls, VPNs, IPS (Intrusion Prevention Systems), and IDS (Intrusion Detection Systems) but also makes a long-term commitment to full-lifecycle security governance capabilities.
Currently, many enterprises' cognition of OT security still remains at the level of deploying security devices such as firewalls and industrial gateways, which appears inadequate to cope with the increasingly complex attack posture in the AI era. Full-lifecycle security governance is becoming the core standard for industrial network security.
Currently, Moxa has obtained the dual certification for IEC 62443-4-1 Maturity Level 3 (ML3) (IECEE and ISCI/ISASecure). This certification covers product security development lifecycle governance, threat modeling, standardized operation of the Product Security Incident Response Team (PSIRT), vulnerability and patch management, as well as supply chain security control.
Regarding how manufacturing enterprises should build the security capabilities of their network foundations, Zhang Heng provided a clear practical path:
01. First, establish the concept of "security by design," incorporating cybersecurity risks into the product design, development, testing, and validation processes;
02. Second, establish a continuous vulnerability management and response mechanism, balancing patch updates with production continuity;
03. Third, pay attention to the supplier's software supply chain management capabilities to ensure that the affected scope can be quickly located when vulnerabilities occur in third-party components;
04. Finally, establish a security governance system that covers the entire product lifecycle.
For enterprises promoting AI-empowered production line upgrades, Zhang Heng also gave practical advice: "First, build a good overall industrial backbone network and establish a reliable, stable, deterministic, and resilient industrial communication foundation. Only with a good network foundation can the effectiveness of AI be sustainable and continuously advanced."
Notably, Moxa will soon appear at the 2026 CIIF (China International Industry Fair) as a "Reliable Communication Expert and Professional Global Expansion Partner," concentrating on showcasing its capabilities in high-end manufacturing and highly reliable deterministic backbone networks, as well as full-scenario real-time edge connection technologies.
It is revealed that Moxa will present five core directions: end-to-end edge access and data uplink solutions, a multi-redundant network resilience system, security foundation capabilities based on the IEC 62443 standard, a visualized panoramic O&M system, and compliance design and global service capabilities for enterprises' global expansion.
These five directions cover the full scenarios of edge access, resilient operation, security, O&M, and global expansion compliance, precisely responding to the core needs of enterprises' intelligent upgrades.
Currently, the intelligent transformation of the manufacturing industry has entered the deep-water zone, and the value of industrial networks is being re-recognized by the entire industry. From connectivity pipelines to AI foundations, every upgrade of industrial networks is paving the way for the deepened implementation of AI applications. Moxa's judgment of "network first" provides a clear underlying logic for manufacturing enterprises exploring intelligent transformation. The development of industrial intelligence cannot ignore the construction of underlying infrastructure. Only by building a solid industrial communication foundation can AI technology release lasting value at the production site.