PaperlayerView exact revision on arXiv

Source-linked AI summary

Artificial Intelligence-Driven Customized Manufacturing Factory: Key Technologies, Applications, and Challenges

Jiafu Wan, Xiaomin Li, Hong-Ning Dai, Andrew Kusiak, Miguel Martínez-García, Di Li

arXiv:2108.03383v2cs.AIcs.MAcs.RO

TL;DR

Traditional mass production lacks flexibility for personalized, small-batch products, motivating AI-driven customized manufacturing. The paper presents and reviews an AI-assisted CM architecture, validates it through packaging and related manufacturing scenarios, and reports the possibility of higher flexibility and efficiency while identifying deployment and technology-transfer challenges.

  • Problem

    Traditional mass production does not flexibly satisfy individual customer requirements, motivating customized manufacturing for personalized products and services.

  • Method

    The paper develops an AI-assisted CM architecture, reviews enabling technologies, and examines intelligent devices, multi-agent coordination, dynamic reconfiguration, scheduling, and deployment.

  • Results

    AI-driven CM supports improved production efficiency and product quality, while the paper validates key technologies in an industrial packaging scenario.

  • Takeaways & Limitations

    AI-assisted customized manufacturing offers the possibility of greater production flexibility and efficiency through adaptive, interconnected, and intelligent manufacturing resources.

Abstract

from arXiv · show

The traditional production paradigm of large batch production does not offer flexibility towards satisfying the requirements of individual customers. A new generation of smart factories is expected to support new multi-variety and small-batch customized production modes. For that, Artificial Intelligence (AI) is enabling higher value-added manufacturing by accelerating the integration of manufacturing and information communication technologies, including computing, communication, and control. The characteristics of a customized smart factory are to include self-perception, operations optimization, dynamic reconfiguration, and intelligent decision-making. The AI technologies will allow manufacturing systems to perceive the environment, adapt to external needs, and extract the processed knowledge, including business models, such as intelligent production, networked collaboration, and extended service models. This paper focuses on the implementation of AI in customized manufacturing (CM). The architecture of an AI-driven customized smart factory is presented. Details of intelligent manufacturing devices, intelligent information interaction, and the construction of a flexible manufacturing line are showcased. The state-of-the-art AI technologies of potential use in CM, i.e., machine learning, multi-agent systems, Internet of Things, big data, and cloud-edge computing are surveyed. The AI-enabled technologies in a customized smart factory are validated with a case study of customized packaging. The experimental results have demonstrated that the AI-assisted CM offers the possibility of higher production flexibility and efficiency. Challenges and solutions related to AI in CM are also discussed.

I. INTRODUCTION

Customized manufacturing addresses personalized products and services, but requires flexible, intelligent production beyond traditional mass production. The paper develops an AI-assisted customized-manufacturing architecture, reviews enabling technologies, and discusses implementation challenges.

  • Customized manufacturing provides personalized products and services as a value-added smart-manufacturing paradigm.
  • The paper develops an AI-assisted CM architecture by merging smart devices, industrial networks, and big-data analysis.
  • The paper reviews state-of-the-art AI technologies and validates key AI-enabled CM technologies using a customized candy-packaging prototype.
  • The paper discusses challenges and possible solutions associated with introducing AI into customized manufacturing.
  • Traditional mass production cannot adapt adequately to rapidly changing personalized-product requirements, motivating customer-to-manufacture systems.
  • Customized smart factories require smart interconnectivity, dynamic reconfiguration, massive-data processing, and deep integration across physical and information environments.

B. Overview of AI technologies

AI technologies span learning, perception, decision-making, and application systems that can support customized manufacturing across design, production, maintenance, and services. Their adoption can improve manufacturing capabilities, but requires substantial data, computing resources, and attention to privacy and latency.

  • AI includes perception, machine learning, deep learning, reinforcement learning, decision making, and applications such as computer vision and intelligent robots.
  • Deep learning combines feature engineering with learning and benefits from massive data, advanced hardware, and algorithms such as CNNs and LSTMs.
  • AI can support customized product design, manufacturing, maintenance, customer management, logistics, after-sales service, and market analysis.
  • Deep-learning adoption is constrained by large data and computing requirements and by poor interpretability in critical industrial tasks.
  • AI-driven CM can improve production efficiency and product quality, facilitate predictive maintenance, and support smart supply chains.
  • Cloud computing can provide intensive processing, but remote data transfer risks confidential-data leakage and high latency for time-sensitive tasks.

III. ARCHITECTURE OF AN AI-ASSISTED CUSTOMIZED MANUFACTURING FACTORY

The AI-assisted customized manufacturing architecture integrates devices, intelligent interaction, AI, and services across cloud, edge, and local computing paradigms. It is designed to support real-time control, adaptive network communication, and intelligent manufacturing services.

  • The framework integrates smart manufacturing devices, intelligent information interaction, AI technologies, and smart services.
  • Cloud intelligence handles comprehensive, time-insensitive analysis and decisions, while edge and local intelligence serve context- or time-aware environments.
  • The device layer provides the physical layer and must meet real-time requirements, with machine-learning algorithms deployable on low-power devices such as FPGAs.
  • The smart interaction layer links the device, AI, and services layers through network devices, controllers, communication protocols, and information storage.
  • AI can predict wireless channels, optimize mobile-network handoffs, and control congestion using recurrent neural networks or reservoir computing.

3) The AI layer:

The AI layer combines cloud, edge, and multi-agent computing to support customized manufacturing tasks. Its applications include data analysis, personalized product recognition, collaborative processing, load balancing, and scheduling optimization.

  • AI algorithms are distributed across computing paradigms: cloud servers can train deep-learning models, while edge servers execute trained models for specific manufacturing tasks.
  • The AI-assisted framework addresses smart devices, interaction, AI technologies, and services while considering edge computing, software-defined networks, and advanced AI.
  • Multi-agent systems support collaborative manufacturing because individual agents are limited to relatively simple tasks, whereas customized production can require complex image-based recognition.
  • Edge servers can run CNN, R-CNN, Fast R-CNN, Faster R-CNN, YOLO, or SSD algorithms and rapidly transfer personalized-product identification results to devices.
  • Edge computing enables quantitative energy-aware models for load balancing, collaborative complex-task processing, and scheduling optimization, supporting production-line logistics, flexibility, and efficiency.

B. Manufacturing resource description based on ontology

Ontology-based manufacturing resource description models devices, attributes, functions, and constraints to support dynamic resource reconfiguration in customized production. Deep-learning forecasts of working-state time slots can inform reconfiguration decisions.

  • Customized manufacturing requires dynamic configuration of distributed resources to support resource sharing, market responsiveness, lead-time optimization, and manufacturing quality.
  • The ontology-based integration framework organizes manufacturing resources through data, rule, knowledge, and resource layers.
  • Ontology modeling describes intelligent production-line devices and attributes by mapping manufacturing resources to functions with associated attributes.
  • A deep-learning algorithm can forecast working-state time slots, which serve as important constraint information for manufacturing-resource reconfiguration.

C. Edge Computing in Intelligent Sensing

Edge computing supports adaptive intelligent sensing by placing processing near manufacturing devices and connecting sensor nodes with edge servers and remote data centers. The section also identifies interoperability and network-resource adjustment challenges.

  • Existing manufacturing-environment sensing is often inflexible because parameters are static or lack prediction functions for advance adjustment.
  • Edge-AI intelligent sensing uses sensor nodes and nearby edge nodes to convert physical conditions into data, preprocess and store it, and provide AI-based computing services.
  • Sensing parameters can be adjusted according to application requirements and task priority, with machine-learning classification supporting rapid responses for high-priority systems.
  • Massive and heterogeneous sensing data motivate AI-supported delivery mechanisms in which sensor nodes both sense and transmit environmental data.
  • Retrofitting legacy equipment with sensors or IoT nodes can improve connectivity for monitoring, but is less suitable for active control and cannot monitor all internal physical quantities.
  • Customized manufacturing information exchange still requires dynamically adjustable network resources and end-to-end data delivery.

A. Software-defined industrial networks

Software-defined industrial networks separate control from data handling and use AI to adapt communication resources to customized manufacturing tasks. The approach supports flexible information sharing, lower adjustment costs, and improved production efficiency.

  • Industrial networks connect customized manufacturing equipment and devices while supporting edge or cloud computing for intelligent information sharing.
  • Software-defined networking separates industrial-network data and control planes, enabling flexible and efficient control in dynamic customized-manufacturing environments.
  • AI-enhanced software-defined industrial networks use network-state data to generate optimized instructions for different devices and customized tasks.
  • Optimization algorithms adjust data-transfer strategies against latency and energy-consumption requirements, reducing the cost of dynamically reconfiguring network resources.
  • End-to-end communication provides lower latency and higher reliability than centralized communication, supporting fully connected systems with differing real-time constraints.
  • AI can improve communication efficiency across physical, MAC, and routing layers by adapting media, access protocols, and routing to device and workload conditions.

VI. FLEXIBLE MANUFACTURING LINE

The flexible manufacturing line combines cooperative agents, AI-based task decomposition, ontology reasoning, and dynamic resource selection to support customized production. These mechanisms coordinate subtasks, reconstruct production functions, and monitor execution across heterogeneous devices.

  • VI. FLEXIBLE MANUFACTURING LINE: Flexible production lines use collective, autonomous, and cross-media reasoning intelligence to support cooperative operation, dynamic reconfiguration, and self-organizing scheduling.
  • VI. FLEXIBLE MANUFACTURING LINE: Multiple agents form collaborative groups for customized tasks, with edge computing supporting cooperative operation and AI-based coordination.
  • VI. FLEXIBLE MANUFACTURING LINE: AI task-decomposition algorithms convert customer product orders into device working procedures under product-manufacturing-time constraints.
  • VI. FLEXIBLE MANUFACTURING LINE: Suitable-size cooperative subgroups can be constructed using cognitive approaches such as the ACT-R model for complicated customized-manufacturing tasks.
  • VI. FLEXIBLE MANUFACTURING LINE: Device agents send status data to edge servers during production, allowing system-wide monitoring while multiple agents execute manufacturing tasks.
  • VI. FLEXIBLE MANUFACTURING LINE: Heterogeneous device capabilities and dedicated-line limitations motivate reconfigurable manufacturing for customized production.
  • VI. FLEXIBLE MANUFACTURING LINE: Ontology reasoning selects devices by function and manufacturing constraints, constructs customized production lines, and enables optimal process planning and functional reconstruction.

C. Self-organizing Schedules of Multiple Production Tasks

Self-organizing schedules address uncertain order arrivals and shared resources by decomposing tasks into steps and time slots, then assigning idle devices through multiple agents. The approach supports simultaneous flexible production but retains humans in the loop.

  • C. Self-organizing Schedules of Multiple Production Tasks: Stochastic and intermittent order arrivals can require production resources to be shared among multiple tasks, motivating self-organizing schedules with time slots.
  • C. Self-organizing Schedules of Multiple Production Tasks: AI-based scheduling divides new production tasks into steps and decomposes producing periods into time slots of different lengths.
  • C. Self-organizing Schedules of Multiple Production Tasks: Edge agents select idle device agents by comparing working-step requirements with device mappings, after which multiple agents self-organize the new-task schedule.
  • C. Self-organizing Schedules of Multiple Production Tasks: Self-organizing schedules can complete simultaneous production tasks using flexible production and improve production-line efficiency.
  • C. Self-organizing Schedules of Multiple Production Tasks: The scheduling approach reduces unnecessary human resource consumption, but appropriate human intervention remains necessary when full automation is unavailable or partial.

A. Prototype platform construction

The prototype customized candy-wrapping line integrates devices, industrial networking, hybrid cloud-edge computing, AI services, and customer-facing ordering. Its preventive-maintenance pipeline collects and analyzes manufacturing data to support monitoring, prediction, and maintenance management.

  • A. Prototype platform construction: The prototype customized candy-wrapping line integrates CM devices, an industrial network, a conveyor, and a cyber-physical system through OPC UA and DDS.
  • A. Prototype platform construction: Its device layer contains five robots, two AGVs, a conveyor system, and a warehouse for material handling and product loading and unloading.
  • A. Prototype platform construction: The industrial network layer connects communication technologies and provides three sub-networks for different latency communication functions.
  • A. Prototype platform construction: Hybrid cloud-edge computing assigns real-time tasks to edge computing and time-insensitive tasks, including historical-data processing, to cloud computing.
  • A. Prototype platform construction: The cloud service layer provides pattern recognition, modeling, knowledge discovery, reasoning, and decision-making capabilities.
  • A. Prototype platform construction: Customers configure candy color, taste, quantity, and variety through an AI recommender, which sends order parameters to the manufacturing cloud.
  • A. Prototype platform construction: The preventive-maintenance pipeline collects, processes, and mines manufacturing data with AI to support monitoring, prediction, and maintenance scheduling.
  • A. Prototype platform construction: AI-based big-data analysis can construct a maintenance knowledge library and decrease smart-manufacturing operation and maintenance management costs.

C. Cloud-assisted customization service

Cloud-assisted customization services integrate cloud computing with customer participation across the production life cycle. They support user-centric, demand-driven customization, but excessive interactions can increase costs and production time.

  • Cloud-assisted customization services integrate cloud computing with customization services to provide user-centric, demand-driven, and service-oriented customer experiences.
  • The service spans early, middle, and later production stages, including recommendation, customer design, remote production monitoring, and logistics information delivery.
  • Customers can participate in production through industrial cloud computing and cloud-assisted technologies, forming a new production model centered on customization services.
  • Data-driven decision-making algorithms such as shape mining, Apriori, and C5.0 support market prediction and rule extraction from manufacturing databases.
  • The customized candy-wrapping platform supports on-demand production, adapts to market changes, and provides a proof of concept for a seamless customized manufacturing pipeline.
  • Discussion: More customization interactions may improve user experience but can also increase expenditure and prolong production time, motivating predefined samples and tutorials.

VIII. CHALLENGES AND ADVANCES

AI, IoT, software-defined networking, and other ICTs can increase customized manufacturing flexibility, intelligence, and efficiency while introducing technical challenges. Current research remains more mature for isolated device intelligence than for generic, integrated smart manufacturing.

  • The convergence of AI, IoT, SDN, and other ICTs increases the flexibility, intelligence, and efficiency of customized manufacturing systems while creating new challenges.
  • Customized manufacturing equipment must support sensing, storage, inference, information interaction, self-diagnostics, hybrid computing, and preventive maintenance.
  • Generic AI-assisted manufacturing remains constrained by the need for further progress in AI and AI-robotics integration, although customized manufacturing is realizable in restricted environments.
  • Existing studies emphasize single-device intelligence, leaving a substantial gap between realistic smart devices and current solutions for integrated manufacturing.

B. Information interaction in CM

Customized manufacturing requires efficient information interaction and dynamic resource reorganization across heterogeneous systems. The paper discusses hybrid networks, AI methods, and deployment constraints, including cost, privacy, workforce, and technology-transfer barriers.

  • Information interaction: Heterogeneous customized manufacturing components require effective connections and highly efficient information interactions.
  • Information interaction: A hybrid industrial network combining modern communications with AI is proposed as a promising solution for application-specific information flows.
  • Dynamic reconfiguration of manufacturing resources: Dynamic reorganization is challenging because customized manufacturing uses small batches, short cycles, flexible production, varied processing crafts, and interacting subsystems.
  • Dynamic reconfiguration of manufacturing resources: Hybrid AI combining knowledge reasoning with swarm intelligence, alongside ML-based process optimization, is suggested for reorganizing manufacturing resources across scenarios.
  • Practical deployment and knowledge transfer: Practical adoption is constrained by upgrade, digitization, computing, AI-service, operating, and personnel costs, especially for smaller enterprises.
  • Practical deployment and knowledge transfer: Cloud or MLaaS outsourcing can reduce computing burdens for SMEs but increases security and privacy risks requiring protection schemes.
  • Practical deployment and knowledge transfer: Technology transfer is limited by non-technical factors and a readiness gap between research feasibility at TRL 1–3 and enterprise requirements at TRL 7–8 or higher.
  • Conclusion: The proposed architectures integrate IoT, edge intelligence, and cloud computing and are validated in an industrial packaging scenario.
Loading 2108.03383v2…