202310-thumbnail_cover story
Revolutionising new industrialisation: the role of big data and AI in driving advancements

Revolutionising new industrialisation: the role of big data and AI in driving advancements

By Hong Kong Industrial Artificial Intelligence and Robotics Centre (FLAIR)*

*FLAIR was solely founded by the Hong Kong Productivity Council.

 

If you choose to listen to this article, you are welcome to download the PDF version of the Journal (December 2023 issue) and activate the "Read Out Loud" function in Adobe Reader. For more details, please read the user's note.

 

This article explores the power of industrial artificial intelligence and industrial big data in advancing new industrialisation. Through presenting the innovative and impactful technologies or platforms being developed in Hong Kong Industrial Artificial Intelligence and Robotics Centre (FLAIR), the article prospects the applications of these newly developed technologies, and inspires the engineers to explore intelligent manufacturing and smart city in Hong Kong.

Industrial AI and industrial big data: technology development and prospection

 

How big data and artificial intelligence (AI) can help improve manufacturing performance? Intelligent manufacturing and smart machines have undergone significant evolution in modern industry due to the support of artificial intelligence and big data approaches. The integration of AI and big data has enabled the development of more sophisticated and intelligent machines, which can operate autonomously with minimal human intervention while delivering more efficient and effective results. As a result, many industries have recognised the potential of AI and have started using it to a greater extent to address challenges and transform themselves towards new industrialisation, a phase characterised by advanced technology, automation, and data-driven decision-making.

 

AI technology is utilised in reducing the chances of machine failure, improving product quality control, increasing productivity, and substantially decreasing product costs, thereby bolstering the potential user base in the market. Predictive quality and yield enable manufacturers to ensure quality products through AI-driven systems that can detect defects early in the production process. Predictive maintenance utilises AI to detect potential problems with machines before they occur, reducing downtime and improving machine reliability. Smart manufacturing monitors physical processes through sensors and other devices, allowing manufacturers to optimise production processes and enhance efficiency. AI in robotics for industrial applications enables the development of more intelligent and sophisticated machines that can operate autonomously. Generative design utilises AI to create multiple design options based on specific parameters, allowing manufacturers to identify the best design.

 

To unlock the full potential of AI in industrial applications, several technical approaches are utilised. These include statistical AI techniques such as support vector machines, decision trees, logistic regression for material classification, random forest, and logistic regression for monitoring and analysing system performance. Deep learning is another approach utilised in industrial applications, and reinforcement and federated learning are also commonly used.

 

In terms of data aspects, AI can be applied to descriptive, predictive, and prescriptive analytics. Descriptive analytics involves analysing historical data to understand past events and trends. Predictive analytics uses AI to anticipate future outcomes based on historical data, while prescriptive analytics utilises AI to determine the most appropriate course of action for a specific goal.


To fully exploit the potential of data generated in the smart industry, big data and AI should be integrated, as data-driven AI algorithms require vast amounts of high-quality data. Big data enable the structuring, integration, and identification of useful patterns in data, which can be used to train AI models to provide outstanding prediction and decision-making capabilities.


Despite the outstanding performance of AI and big data, several challenges hinder their successful deployment in new industrialisation. These challenges include few-shot learning, imitation learning, scalability issues, machine-tomachine communication, interpretation, and human-robot collaboration. Addressing these challenges could pave the way for potential future research directions that enable the full realisation of AI and big data's potential in new industrialisation.


Overall, the integration of AI and big data has revolutionised the industrial landscape, enabling manufacturers to optimise their operations, reduce costs, and deliver high-quality products to the market. Therefore, it is no surprise that AI and big data are being increasingly adopted by various industries to stay competitive in an increasingly data-driven world.

 

Impacts on Hong Kong industry - artificial intelligence and robotics in new industrialisation


In Hong Kong, the adoption of industrial AI and industrial big data technologies is on the rise, particularly in the manufacturing sector. The HKSAR Government has recognised the importance of these technologies and supports companies looking to adopt them. The development and adoption of Industrial AI and industrial big data technologies offer significant opportunities for Hong Kong’s industry to stay competitive in an increasingly data-driven world. By harnessing the power of these technologies, Hong Kong's industry can optimise its operations, reduce costs, and deliver high-quality products to the market while remaining competitive globally.

 

The idea of new industrialisation was first highlighted in the Hong Kong Innovation and Technology (I&T) Development Blueprint, which refers specifically to the type of industrialisation which is driven by informatisation and capable of achieving developments in leaps and bounds and strengthening sustainable development. With new industrialisation driving the transformation of manufacturing processes, the physical world of production is evolving towards a higher level of automation. This is achieved through the merging of the physical world with the digital world of information technology including industrial AI and robotics, Internet of Things (IoT), big data, virtual reality (VR) and augmented reality (AR), resulting in the creation of cyber-physical systems. This convergence is expected to significantly impact a vast range of industries, and the potential for growth is exponential.


To achieve new industrialisation in Hong Kong and enable it to capitalise on this opportunity, the Hong Kong Productivity Council (HKPC) established the Hong Kong Industrial Artificial Intelligence and Robotics Centre, also known as FLAIR. The centre focuses on research and development in the field of industrial AI and robotics and works closely with industrial partners to optimise their business performance.

 

FLAIR focuses on research and development in the field of industrial AI and robotics and works closely with industrial partners to optimise their business performance.

 

FLAIR aims to serve as a platform for collaboration between industry and academia, bringing together experts from various fields to share knowledge and insights. Through such collaboration, FLAIR seeks to advance the development of intelligent manufacturing and smart city technologies in Hong Kong, thereby contributing to the city’s overall economic growth and competitiveness.


Overall, FLAIR's focus on industrial AI and robotics is a critical component of Hong Kong's new industrialisation efforts. By embracing the opportunities presented by new industrialisation and developing cutting-edge technologies, Hong Kong can establish itself as a leader in intelligent manufacturing and smart city development, thereby enhancing its economic competitiveness and securing its position as a global innovation hub.

 

Future of new industrialisation in Hong Kong


The HKSAR Government is actively promoting new industrialisation and advanced manufacturing based on new technologies and smart production. This promotion has led to an increase in demand for research and development in Hong Kong. The future of new industrialisation in Hong Kong is promising, with several key factors driving its growth. The government has allocated significant resources to the industry’s development support, including funding for research and development, tax incentives for manufacturers, and the establishment of several industrial parks. In response to the Hong Kong I&T Development Blueprint, FLAIR is developing projects to facilitate new industrialisation in the city.

 

These projects aim to enhance the competitiveness of Hong Kong’s manufacturing industry through the development of innovative technologies. By leveraging cutting-edge technologies such as AI, robotics, and IoT, FLAIR seeks to enable local manufacturers to optimise their production processes and improve their overall efficiency.

 

Moreover, FLAIR’s efforts are expected to have a significant impact on Hong Kong's economy, creating new job opportunities and enhancing the city's global competitiveness. By supporting the development of advanced manufacturing technologies, Hong Kong can establish itself as a leader in the field and will attract investment from around the world.

 

Overall, the government's promotion of new industrialisation and the development of advanced manufacturing is a crucial element in Hong Kong's economic development strategy. By working with organisations such as FLAIR, the city can leverage the latest technologies to drive growth and create a more prosperous future for all.

 

About FLAIR


HKPC and the RWTH Aachen Campus from Germany join forces to establish the Hong Kong Industrial Artificial Intelligence and Robotics Centre (FLAIR). Located at the Hong Kong Science Park, FLAIR is part of AIR@InnoHK, which is one of the two world-class research clusters being established by InnoHK Clusters of the HKSAR Government to support Hong Kong’s development into an international innovation and technology hub.

 

As the sole founder, HKPC will work together with RWTH Aachen Campus, the major collaborator of FLAIR, by bundling together their strengths and repertoire of resources in technical expertise, networks, innovation, and patents etc. to support FLAIR’s operation.


Riding on the network of HKPC, FLAIR has identified global collaboration partners in academic and research and development institutions, including Tsing Hua University and The Hong Kong University of Science and Technology (HKUST).


Vision

 

  • Be the pioneer of the research and development and applications of AI and Robotics technologies in Hong Kong

 

Missions

 

  • Develop innovative and impactful technologies or platforms on AI and robotics to facilitate intelligent manufacturing for enterprises
  • Collaborate with enterprises and institutes around the Greater Bay Area (GBA) to promote intelligent manufacturing cooperation

 

Research and development technologies developed at FLAIR

 

AI and big data technologies can enhance various manufacturing processes, and FLAIR’s expertise encompasses a wide range of aspects within this realm, including but not limited to:

 

  1. Digital shadows for distributed manufacturing:
    a) Big data based intelligent production planning platform
    b) Industrial equipment prognosis and health management
    c) AI-based monitoring
  2. Flexible production
    a) Flexible assembly line (details shown below)
    b) Line-less mobile production
    c) AI monitor and decision support (details shown below)
    d) Multi-purpose mobile robotic platform
  3. Intelligent automation for manufacturing
    a) Robotics system with vision for sorting application
    b) Interactive control system in assembly application
    c) Universal gripper for food industry

 

 

Self Photos / Files - 擷取1AI and big data technologies can enhance various manufacturing processes and
FLAIR’s expertise encompasses a wide range of aspects within this realm

 

 

Highlighted technologies being developed in FLAIR

 

First example: Bottleneck identification and prediction in the air cargo terminal logistics processLogistics is a critical aspect of many businesses today. It involves the management of the flow of goods and services from the point of origin to the point of consumption. Logistics processes are often complex and involve multiple stakeholders, which can lead to inefficiencies and bottlenecks. As a significant part of global transportation, the logistics process at air cargo terminals is a complex and intricate operation that requires efficient handling of various cargo types, including perishable goods and hazardous materials. Fluid operations of warehouses and distribution centres are essential to support the international exchange of materials. However, these facilities are often plagued with equipment faults, systematic problems, and human factors. It is essential to identify bottlenecks in the logistics process to optimise efficiency, reduce costs, and improve customer satisfaction.


Identifying bottlenecks
Bottlenecks are a common problem in logistics processes. They occur when there is a constraint in the process that limits the overall throughput. For example, a warehouse with limited capacity can become a bottleneck if it cannot keep up with the demand for goods. Bottlenecks can cause delays, increase costs, and reduce logistics efficiency. Air cargo terminals provide express cargos and courier operators with their own sorting facilities and can process millions of tonnes of cargo a year. Specialist facilities allow the centre to handle anything from livestock to precious items, including diamonds, cash, and gold bullion. Identifying bottlenecks in air cargo terminal logistics processes is critical to optimising the overall process.


One approach to identifying bottlenecks in air cargo terminal logistics processes is to use data-driven methods, which involve the collection and analysis of data from various sources, including event logs, sensor data, and historical data. By using data-driven methods, stakeholders can gain insights into the performance of the logistics process and identify areas for improvement. Currently, radio frequency identification (RFID) tags can provide real-time tracking and monitoring of cargo movements throughout the logistics process. The tags can be attached to cargo containers, and their movement can be tracked and monitored in real-time. By analysing such movement data, stakeholders can identify bottlenecks in the logistics process, such as delays in the handling of cargo, and take corrective action to improve the overall process.

 

A typical data-driven method for identifying process bottlenecks is process mining, which involves the extraction of knowledge from event logs. Event logs are records of the activities within an organisation, and they can provide valuable insights into the efficiency and effectiveness of business processes. Process metrics are quantitative measures of a process, and they can be used to identify bottlenecks. By analysing process metrics, logistics companies can identify areas of the process that are slow or inefficient and take steps to improve them. By using process mining, logistics companies can gain a better understanding of their processes and identify areas for improvement.


Another approach to identifying bottlenecks in air cargo terminal logistics processes is process mapping. Process mapping involves creating a visual representation of the logistics process, including the various steps and stakeholders involved in the process. By analysing the process map, stakeholders can identify areas of the logistics process where bottlenecks are likely to occur.

 

Bottleneck prediction

Bottleneck prediction is also a critical aspect of air cargo terminal logistics process management. By predicting bottlenecks, stakeholders can take proactive measures to prevent them from occurring. Predicting bottlenecks involves analysing data and identifying patterns that can indicate the
potential for bottlenecks.


One approach to predicting bottlenecks in air cargo terminal logistics processes is to use machine learning algorithms. Machine learning algorithms can be trained on historical data to predict future bottlenecks in logistics. By analysing data from previous logistics processes, machine learning algorithms identify patterns that can indicate bottleneck potential. For example, machine learning algorithms can be used to analyse the historical data on cargo movements and identify patterns that indicate the potential for delays in the logistics process. This could include delays in customs clearance, cargo handling delays, or delays in cargo arrival or departure.

 

Another approach to predicting bottlenecks in air cargo terminal logistics processes is to use simulation models. Simulation models involve creating a digital model of the logistics process and simulating various scenarios to identify potential bottlenecks. For example, simulation models can be used to simulate different scenarios, such as an increase in the volume of cargo, changes in the cargo mix, or changes in the arrival or departure schedules of cargo. By simulating these scenarios, stakeholders can identify potential bottlenecks and take corrective action to prevent them from occurring.

 

Moreover, real-time bottleneck prediction involves analysing event logs as they occur and using this information to predict potential bottlenecks. This approach can be particularly useful in fast-paced logistics environments, where bottlenecks can develop quickly and unexpectedly. By predicting bottlenecks in real-time, logistics companies can take immediate preventive action.

 

The Logistics Process Intelligent System

To achieve the goal of bottleneck identification and prediction for air cargo terminals and logistics distribution centres, FLAIR developed the Logistics Process Intelligent System (the System). The System originated as an automatic solution developed to identify problems in complicated processes. This vision has soon expanded to a more generalised solution, suitable for any conveyor systems.

 

The System aspires to boost processing ability in conveyor systems by applying data mining and deep learning techniques to log data. It takes root in a fully automatic workflow, from ingesting raw data to data processing using AI, and finally displaying results on a well-designed interactive dashboard.

 

To achieve the goal of bottleneck identification and prediction for air cargo terminals and logistics distribution centres, FLAIR developed the Logistics Process Intelligent System, which began with developing an automatic solution to identify such problems in the complicated processes.

 

Existing solutions are highly modularised, often serving only one specific purpose. The System provides a unique one-stop solution for any industry that operates on a conveyor system, handling all stages from raw data ingestion to data processing by algorithms, and ultimately displaying on an interactive dashboard. Furthermore, the System is being expanded to a more generalised system can be widely applied to any conveyor system.


Users can identify bottlenecks to accelerate process blockage clearance; monitor complicated processes by evaluating package dwell time and throughput of each station; identify faults to speed up equipment recovery; and trace the performance of previous processes to prevent recurrent problems.


The first version of the System was developed and delivered specifically for our industrial partner. During the development stage, we closely communicated with the end user to ensure that our product would cater to their specific needs. We also offered workshops for end users to test the product and suggest wish lists of features to enhance the System. To ensure that it could be applied in the wider logistics industry, the System was expanded to include general features for any industries that involve logistics processes.

 

 

Self Photos / Files - 擷取2The Logistics Process Intelligent System: equipment throughput

 

Self Photos / Files - 擷取3The Logistics Process Intelligent System: ULD dwell time

 

 

Maximising efficiencies for logistics processes
The immediate impact of the System is to empower industries that involve conveyor belt systems such as warehouses, distribution centres, and airports by maximising order processing efficiency and minimising lead time in waiting for orders. Through active monitoring by the System, chokepoints can be easily identified. The System also advises on resolving or alleviating the negative impacts brought by such bottlenecks.


In longer term, the System helps create more efficient logistics systems for the society. It helps conveyor belt facilities reduce operating costs while end users and consumers enjoy a shorter waiting time for goods to arrive at their doorsteps.


Recognitions


The System has been recognised, including being admitted into an ideation program sponsored by Hong Kong Science and Technology Parks (HKSTP). It has been granted Hong Kong short term patent (No.: HK30081800) and is pending for the Chinese Patent.


Second example: automated detection of surface defects with a few object instances
In production lines, defect inspection still relies mainly on human efforts. Workers need to check carefully from all angles to ensure the product quality. With the continuous increase in labour costs and the rapid development of automated production lines, vision-based industrial automated defect detection has been widely used in various industries, such as computer, communication, and consumer electronics (3C), automobiles, home appliances, machinery manufacturing, semiconductor, and electronics industries. In the next generation of manufacturing, automatic defect detection will be playing an increasingly important role. To accelerate enterprises’ transformation and overhaul, establish industry standards, and realise modern flexible production, a novel vision-based industrial auto defect detection solution has been developed which bears far-reaching significance and broad market prospects. At present, the technical challenges of vision-based industrial auto defect detection lie in the following aspects:

 

Flexible system

Market demands dictate the current trend in the manufacturing industry, which is to migrate to high-mix low-volume (HMLV) manufacturing. Since this trend requires more product variations but fewer quantities for each variation, the production lines need to change their quality control setup much more frequently. It is difficult for vision-based industrial hardware system with fixed working distance and angle to detect defects in objects of various sizes and shapes. When the products change, experienced engineers are needed to re-adjust the hardware system or even re-design the hardware system.


Detection of irregular and reflective objects

In cases of objects with reflective surfaces, one of the solutions is to use powder to change the reflective properties of the objects and eliminate the impact of reflection on detection. Spraying powder is time-consuming, and tiny defects will be covered with powder. The process of cleaning the powder also adds to the cost. In cases of irregular objects, the image capturing angle needs to be manually set by experienced engineers, which adds to the difficulty of meeting frequent product lines changes and flexible production.


Long software deployment period

Supervised deep learning is used in the industrial defect detection software system with the rapid development of machine learning. The procedure can be summarised as collecting image data, manual labelling, training model, and deployment verification. The above steps need to be repeated each time when the product changes. Among  them, image data collection and manual labelling are significantly time-consuming. Since production lines involve imbalanced dataset with non-defective products significantly outnumbering defective ones, it takes a lot of time to collect product data of different types of defects to train an accurate model. Supervised learning can only conduct model training for known defect types. To ensure image labelling quality, workers have to be trained to use marking tools and understand labelling guidelines. Such training, however, is time-consuming. Software deployment based on supervised learning usually takes a year or more.

 

Solution for irregular, reflective objects

Heeding the current technical difficulties of automatic defect detection in industrial vision, FLAIR is developing the Anomaly Recognition Graphical Universal System (ARGUS), a flexible defect detection system for irregular, reflective objects. Argus is a many-eyed giant in Greek mythology. With more than 100 eyes, he can look every way and keep watch even in sleep. FLAIR’s ARGUS steps even further. ARGUS can not only observe complicated objects from all-around views with nimble “hands” (collaborative robotics) but also detect tiny defects on object surfaces with intelligent “eyes” (high-resolution cameras). ARGUS is equipped with smart “AI brains” that enable it to discover the best views of the products autonomously and to uncover unknown defects after learning from limited positive models.


Based on the hand-eye system, ARGUS combines collaborative robotics with high-resolution industrial-grade optical systems, including 2D cameras and 3D scanners. Its aim is to support data acquisition from objects of different shapes and sizes from multiple angles. With a 3D camera, it can reconstruct a 3D model of any new product variation at once. Collaborative robotics can automatically find the best viewing angles for the entire product on the basis of the 3D model. With a few images from the positive samples captured by the high-resolution 2D camera, ARGUS can quickly train its AI algorithms to detect anomalies on the product surface. 


ARGUS is developed so that the manufacturing industry may adapt itself to the latest market environment. It provides a flexible solution of quality control with self-adaptability, high efficiency, and lower cost. The solution also promotes Industry 5.0. ARGUS, as a flexible quality control solution, has embraced the advantages of both manual and machine inspection. Using its built-in 3D reconstruction algorithm, it can adapt to products with different shapes in the same way as human inspectors can. By adding an in-house AI anomaly detection algorithm, instead of many images with defects, the system only requires a limited number of images from positive sample products for the AI to learn to detect anomalies on the product surface. Combining both of these features and compared with traditional machine inspections, ARGUS boasts a high degree of adaptability. This eliminates the cost of designing and developing new hardware or software for new products in production lines which greatly reduces the overall cost and time for manufacturers when deploying a new product.

 

ARGUS is developed for the manufacturing industry to fit the latest market environment.

It provides a flexible solution for quality control with self-adaptability, high efficiency, and lower cost.

 

 

Self Photos / Files - 擷取4Anomaly-Recognition Graphical Universal System (ARGUS)

 

Self Photos / Files - 擷取5

ARGUS team exhibiting the technology to public

 

Recognitions


The system has been recognised both internationally and locally, including being awarded a Silver Medal at the International Exhibition of Inventions of Geneva 2023 and Outstanding Technology (Building) at the TechConnect Awards 2023, as well as winning a Gold Award at Asia International Innovation Invention Exhibition 2023. The system is granted Hong Kong Patent (No.: HK30012023) and pending Chinese Patent.

 

 

 

Explore Hong Kong Engineer