Source-linked AI summary
Big Data Analytics for Manufacturing Internet of Things: Opportunities, Challenges and Enabling Technologies
Hong-Ning Dai, Hao Wang, Guangquan Xu, Jiafu Wan, Muhammad Imran
TL;DR
Manufacturing IoT generates massive, heterogeneous, real-time data whose analytics can create business value but also poses research challenges. This paper surveys the MIoT analytics life cycle, its necessities and challenges, and enabling technologies across its phases, then outlines future directions. It concludes that big data analytics can help promote manufacturing’s evolution toward smart manufacturing.
Problem
MIoT data’s massive volume, heterogeneous types, and real-time velocity create research challenges despite its potential business value.
Method
The paper surveys MIoT big-data analytics through a three-phase life cycle: acquisition, preprocessing and storage, and analytics.
Results
The paper summarizes enabling technologies for the three life-cycle phases and discusses future directions and open research issues.
Takeaways & Limitations
Big data analytics is presented as an important means of supporting manufacturing’s evolution into smart manufacturing.
Abstract
from arXiv · showhide
The recent advances in information and communication technology (ICT) have promoted the evolution of conventional computer-aided manufacturing industry to smart data-driven manufacturing. Data analytics in massive manufacturing data can extract huge business values while can also result in research challenges due to the heterogeneous data types, enormous volume and real-time velocity of manufacturing data. This paper provides an overview on big data analytics in manufacturing Internet of Things (MIoT). This paper first starts with a discussion on necessities and challenges of big data analytics in manufacturing data of MIoT. Then, the enabling technologies of big data analytics of manufacturing data are surveyed and discussed. Moreover, this paper also outlines the future directions in this promising area.
I. INTRODUCTION
The paper frames MIoT as a smart-manufacturing infrastructure whose massive, heterogeneous, real-time data creates both substantial value and analytics challenges. It surveys these challenges and corresponding enabling technologies while outlining future research directions.
- MIoT connects manufacturing equipment and sensing devices to computing platforms, linking the physical manufacturing environment with cyberspace and decision-making algorithms.
- MIoT data has large volume, heterogeneous structures, and real-time generation, creating challenges across the data-analytics life cycle.
- Existing manufacturing surveys often lack introductions to enabling technologies that correspond to identified analytics challenges.
- The paper summarizes MIoT characteristics and the analytics life cycle, discusses necessities and challenges, surveys technologies for acquisition, preprocessing, and analytics, and outlines future directions.
B. Life cycle of big data analytics for MIoT
The proposed MIoT analytics life cycle proceeds through data acquisition, preprocessing and storage, and data analytics. Analytics supports manufacturing operations, equipment availability, product quality, supply chains, and customer experience.
- Life cycle: The life cycle comprises three consecutive stages: data acquisition, data preprocessing and storage, and data analytics.
- Data acquisition: Data acquisition collects raw manufacturing data from sources such as RFID tags and transmits it to storage through wired or wireless systems.
- Data preprocessing and storage: Preprocessing addresses the volume, redundancy, and uncertainty of raw data through cleaning, integration, and compression before storage.
- Data analytics: Data analytics applies analytical schemes to massive manufacturing datasets to extract valuable information.
- Analytics benefits: Predictive analytics can improve operations, reduce downtime, improve product quality, enhance supply-chain efficiency, and improve customer experience.
D. Challenges of big data analytics for MIoT
MIoT analytics must handle massive, heterogeneous, real-time data across acquisition, preprocessing, storage, and analysis. The resulting challenges include representation, transmission, integration, quality, storage, scalability, efficiency, correlation, mining, privacy, and security.
- Data acquisition: Data acquisition is challenged by representing heterogeneous data and transmitting massive volumes under bandwidth and energy constraints.
- Data preprocessing: Preprocessing must integrate heterogeneous data, reduce temporal and spatial redundancy, clean noisy or erroneous readings, and compress large datasets.
- Data storage: Storage systems must balance reliability, persistency, scalability, efficiency, and cost while supporting massive heterogeneous datasets and concurrent analytics queries.
- Data analytics: Analytics faces difficulties from tremendous volume, heterogeneous structures, high dimensionality, and spatial-temporal correlations.
- Privacy and security: Privacy-preserving analytics must address MIoT’s massive volume, heterogeneous structures, and spatio-temporal correlations because conventional schemes may not apply.
III. ENABLING TECHNOLOGIES
The paper organizes MIoT big-data analytics enabling technologies according to the life cycle’s three phases: acquisition, preprocessing and storage, and analytics.
- Enabling technologies are categorized into data acquisition, data preprocessing and storage, and data analytics.
A. Data acquisition
MIoT data acquisition draws on diverse manufacturing sources and transmits collected data through wired or wireless links. Wireless technologies trade off coverage, bandwidth, power consumption, range, connection capacity, and data rate.
- Manufacturing sources include sensors, RFID or other tags, and data from production lines, warehouses, and other supply-chain sectors.
- Collected manufacturing data is transmitted to storage through either wired or wireless communication systems.Industrial Ethernet is a typical wired connection, while wireless communication avoids wiring and related infrastructure.
- Wireless technologies differ in their coverage and bandwidth capabilities, so MIoT networks combine technologies for varied acquisition requirements.The paper compares technologies including RFID, Bluetooth LE, WBAN, and LPWAN-related systems.
- LPWAN provides wide coverage with lower power consumption, but its low data rate means it should complement conventional wireless technologies.LPWAN ranges from 1km to 10 km; NB-IoT is reported to support up to 52,547 connections and up to 250 kps.
1) Data preprocessing:
MIoT preprocessing and storage address heterogeneous, noisy, redundant data before analytics. The surveyed technologies span cleaning, integration, compression, scalable storage, database management, distributed computing, and lightweight virtualization.
- MIoT data includes structured, semi-structured, and non-structured forms such as sensor readings, RFID data, product records, logs, audio, and video.
- Industrial data can be erroneous or noisy because of collection interference, equipment or sensor failures, and communication loss.Wireless transmission is also affected by blockage, shadowing, fading, and battery depletion.
- Preprocessing approaches include data cleaning, data integration, and data compression to address redundancy, inconsistency, missing values, and measurement errors.The surveyed cleaning mechanisms include duplicate removal, missing-value interpolation, and energy-efficient cleaning.
- MIoT storage uses storage infrastructure and data-management software, including distributed file systems, DBMSs, distributed computing models, virtual machines, and containers.Containers provide lightweight virtualization with faster booting, smaller size, and lower resource consumption than virtual machines.
- MapReduce and its extensions support large-scale processing, while alternatives include Spark, Pregel, Hive, GraphLab, Dryad, and Nephele/PACTs.MapReduce extensions address the lack of iterations or recursions required by many analytics applications.
C. Data analytics
MIoT data analytics comprises statistical modeling, data mining, machine learning, and visualization. These approaches quantify relationships, extract information from massive datasets, and support predictive analysis.
- Typical MIoT analytics approaches include statistical modeling, data mining, machine learning, and data visualization.
- Statistical methods include descriptive statistics, inferential statistics, and stochastic modeling for relationships, generalization, dynamic traffic features, mobility, and object tracking.
- Data mining extracts useful information from massive datasets using algorithms such as Apriori, FP-Growth, DBSCAN, GSP, SPADE, and PrefixSpan.
- Machine learning constructs self-adaptive algorithms that learn from existing data and perform predictive analysis.Examples include SVMs, naive Bayes, decision trees, k-NN, hidden Markov models, Bayesian networks, neural networks, and ensemble methods.
2) Taxonomy of data analytics approaches in MIoT:
The paper organizes MIoT analytics into descriptive, diagnostic, predictive, and prescriptive levels of increasing complexity and extracted value. The levels progress from describing events to diagnosing causes, anticipating futures, and selecting actions.
- The four categories differ in complexity and extracted value, with descriptive and diagnostic analytics characterized as reactive.
- Descriptive analytics: Descriptive analytics explores historical data to explain what happened, reveal characteristics, recognize patterns, and identify relationships.It can be applied across the manufacturing data life cycle, including real-time resource monitoring.
- Diagnostic analytics: Diagnostic analytics examines data to understand event causes, identify equipment faults, detect anomalies, and help reduce machine downtime.Examples use SVMs, artificial neural networks, Kalman filters, supervised learning, and unsupervised learning.
- Predictive analytics: Predictive analytics uses historical data to anticipate future trends, including tool wear, product defects, customer behavior, and device maintenance needs.The surveyed applications include random forests, cost-sensitive decision-tree ensembles, deep learning, and Bayesian networks.
- Prescriptive analytics: Prescriptive analytics extends earlier analyses to recommend decisions for predicted outcomes through simulation, decision-making, optimization, and reinforcement learning.A cited application simulates configuration and procedural training in a bio-ethanol plant.
3) Data visualization in MIoT:
Data visualization helps extract and interpret informative values from complex, high-dimensional MIoT data. The section situates visualization alongside analytics methods and distributed computing infrastructure for MIoT applications.
- Data visualization helps extract and interpret informative values from complex and high-dimensional MIoT data.
- Typical visualization messages represent time series, rankings, frequency distributions, deviations, correlations, part-to-whole relationships, and geographic information.
- Basic techniques include statistical plots, text word clouds, correlation matrices, network diagrams, and heat maps.
- Common visualization toolboxes include Matlab, gnuplot, Seaborn, Pandas, Matplotlib, Tableau, Plotly, and Sisense.
- MIoT prototypes combine production-line devices, edge servers, remote cloud servers, and distributed processing platforms for diverse data tasks.
- The hybrid edge/cloud framework outperforms pure cloud and pure edge schemes for larger image sizes, including 16 MB, 18 MB, and 20 MB.Latency averages use 100 images per image size.
IV. FUTURE RESEARCH DIRECTIONS
Future research directions address privacy and security risks throughout MIoT data acquisition, preprocessing, storage, and analytics. The section identifies encryption, access control, traceability, and privacy-preserving analytics as relevant directions.
- Privacy and security: Privacy concerns proper data utilization while preserving enterprise information, whereas security concerns data confidentiality, integrity, and availability.
- Security assurance in data acquisition: Wireless data acquisition is vulnerable to attacks such as passive eavesdropping because wireless media are open.Encryption is identified as a typical countermeasure, although cryptography may not be feasible in every IoT network.
- Privacy preservation and security assurance in data preprocessing and storage: Distributed MIoT data across manufacturing sites can be vulnerable during preprocessing and storage.Suggested directions include key management, authentication, access control, and traceability of data access or modification.
- Privacy preservation in data analytics: Encrypted MIoT data must often be decrypted before analytics, and decryption can reduce analytics efficiency.The section frames balancing privacy preservation and efficiency as an open challenge.
B. Edge Computing for big data analytics in MIoT
Cloud computing offers manufacturing flexibility but can suffer from latency, bottlenecks, single points of failure, and privacy leakage. Edge computing complements cloud resources, while effective cloud–edge collaboration and lightweight analytics remain open challenges.
- Cloud computing can save ICT capital investment and provide flexible ICT resources to small and medium enterprises, but has several limitations.The listed limitations are high latency, performance bottlenecks, single points of failure, and privacy leakage.
- Hybrid edge/cloud computing can support sensing, monitoring, and controlling near factories and enterprises.The paper’s case study demonstrates the effectiveness of hybrid edge and cloud computing in MIoT.
- Cloud–edge collaboration must allocate computational tasks across resources with different capabilities and upload delays.Remote clouds generally provide greater computing capability, while local edges generally offer shorter upload delays.
- Delay-critical analytics may need local execution, but edge resource limits make conventional methods too complicated.Cloud-trained models can be transferred to edge servers, although model transmission can impose substantial communication cost; AlexNet is cited at 240MB.
C. New data analytics methods for MIoT data
The paper identifies unresolved issues in MIoT analytics, including privacy and security, severe class imbalance, and real-time stream processing. It argues that new methods are needed for large, heterogeneous, and continuously generated manufacturing data.
- Many open research issues remain in developing data analytics methods for MIoT data.
- Imbalanced data samples: Manufacturing datasets can contain extremely imbalanced positive and negative samples, with a reported ratio of 99,000,000 to 1.Conventional analytics methods are challenging to apply to such imbalanced datasets.
- Stream data processing: MIoT generates tremendous volumes of real-time stream data, including sensory data from industrial wireless sensor networks.
- Stream data processing: Entire MIoT streams cannot be stored and processed in computer memory, limiting methods that require the complete dataset in memory.New approaches are needed to process massive MIoT data streams.
- The survey organizes enabling technologies across data acquisition, data preprocessing and storage, and data analytics phases.It also discusses necessities, challenges, future directions, and open research issues.