SRI LANKA INSTITUTE OF INFORMATION TECHNOLOGY

Faculty of Engineering

A Blockchain and Directed Acyclic Graph (Tangle)-Based Framework for

by

Seniru Dilmith Rathnayake

December 2024
The study titled "An Innovative Machine Learning Model for Supply Chain Management"
explores the application of Conditional Generative Adversarial Networks (CGANs) to address the
complexities of dynamic partner selection in supply chains. By integrating key modules such as
order management, inventory control, and transportation planning, it presents a comprehensive
framework to enhance operational efficiency and decision-making. Through CGANs, the study
optimizes procurement processes by refining supplier selection and order allocation using
advanced feature extraction and multi-criteria analysis. In warehousing, predictive analytics
powered by machine learning enable precise demand forecasting, ensuring optimal inventory
levels while minimizing holding costs and stockouts. Real-time traceability is achieved using
RFID-based tracking systems. For distribution, the study implements a Vehicle Routing Problem
with Time Windows (VRPTW) model to streamline delivery schedules and routes, ensuring cost-
effective and timely logistics. Underpinned by the scalable and robust Spring, Struts, and Hibernate
(SSH) framework, this research makes a significant contribution to advancing supply chain
management through machine learning.
The study presents notable advancements in supply chain optimization; however, several critical
aspects remain unaddressed. Transparency mechanisms, including traceability, ethical compliance,
stakeholder accessibility, and trust-building measures, are not explicitly incorporated, limiting the
framework's ability to ensure accountability and collaboration across the supply chain. In
warehousing, while the study emphasizes inventory management and RFID-based tracking, it
lacks specific methodologies for pallet stacking or organizing storage hierarchies based on demand
forecasts, leaving a gap in addressing dynamic inventory prioritization. Additionally, while the
Vehicle Routing Problem with Time Windows (VRPTW) is applied, the study overlooks critical
factors such as vehicle capacity constraints, efficient loading optimization, and the demand
hierarchy of delivery locations, which are essential for cost-effective and prioritized logistics.
Furthermore, although predictive outputs for demand management and scheduling are discussed,
the study does not establish an explicit framework to integrate these predictions with routing
optimization, limiting the practical applicability of its findings.
Transparency in supply chains is crucial for promoting fairness, ethical practices, and
accountability. In the cocoa industry, opaque supply chains result in West African farmers earning
less than 7% of the chocolate’s retail price[1]. Similarly, a lack of transparency in the electronics
sector enables worker exploitation and unsafe conditions in cobalt mining in the Democratic
Republic of Congo[2]. Blockchain technology addresses these issues by providing an immutable
ledger accessible to all stakeholders. Manufacturers benefit from improved traceability,
distributors ensure compliance and fraud prevention, and consumers gain access to verified
product data. Additionally, smart contracts automate trust and reduce reliance on intermediaries,
fostering a more transparent and accountable supply chain ecosystem. The blockchain developed
for the research is heavily influenced by the studies "Blockchain-Powered Security and
Transparency in Supply Chain: Exploring Traceability and Authenticity through Smart Contracts"
and "Transparency by Design for Blockchain-Based Supply Chains." The first study highlights
how blockchain enhances traceability and authenticity through immutable ledgers and smart
contracts, addressing challenges such as data tampering and process accountability, which
informed the use of SHA-256 hashing for secure data verification and Fernet symmetric encryption
for data protection in the developed blockchain[3]. The second study introduces the "Transparency
by Design" framework, emphasizing real-time visibility, traceability, and stakeholder trust, which
directly shaped the blockchain's design by enabling synchronized updates to CSV files for real-
time transparency and the use of smart contracts to automate trust and minimize intermediary
reliance[4]. Furthermore, both studies underscore the importance of decentralized verification
1
mechanisms, which guided the incorporation of tamper-proof structures to maintain system
integrity. The blockchain developed for the research ensures secure, transparent supply chain
management by recording transaction details in an immutable ledger. It uses SHA-256 hashing for
tamper resistance and Fernet symmetric encryption for data security. A genesis block initializes
the chain, with CSV integration providing real-time visibility for stakeholders. An HTTP server
enables decentralized interactions, allowing block additions and chain retrieval. Rooted in
principles from the referenced studies, the blockchain addresses traceability, compliance, and
ethical challenges in modern supply chains.
Traditional blockchain architectures face significant limitations in handling multiuser approvals
and parallel transaction processing due to their inherently linear and sequential structure, leading
to bottlenecks as transactions are processed one at a time. Studies such as Provenance-Aware
Traceability Framework for IoT-Based Supply Chain Systems (ProChain) and A Lightweight
DAG-based Blockchain for Vehicular Social Networks (LDV) highlight these constraints,
emphasizing how they hinder scalability and efficiency in dynamic environments. In contrast,
DAG (Directed Acyclic Graph)-based architectures address these challenges by representing
transactions as vertices in a graph rather than sequential blocks, enabling concurrent validations
and multiuser approvals without delays. ProChain and LDV illustrate how DAG facilitates faster
consensus, improved throughput, and seamless parallel processing. The DAG architecture
developed for procurement builds on these insights from ProChain and LDV to overcome the
limitations of traditional blockchains. Unlike blockchain’s sequential structure, which causes
delays in multiuser approvals and transaction processing, DAG allows transactions to function as
independent vertices linked through hashes, enabling concurrent validations and reducing
bottlenecks. The first study guided the implementation of multiuser approval mechanisms and
immutable transaction storage to enhance traceability and accountability[5], while the second
study informed scalable parallelism and conflict resolution through weighted random walks[6].
Additionally, with real-time CSV-based data exports and efficient conflict handling, DAG offers
faster approvals, seamless parallel operations, and enhanced transparency, making it an ideal
solution for meeting the dynamic demands of procurement compared to blockchain’s linear design.
The DAG (Directed Acyclic Graph) developed for the research represents transactions as
independent vertices linked through hashes, enabling concurrent processing and multiuser
approvals without delays. Each transaction references previous transactions (tips) selected via
weighted random walks, prioritizing high-trust transactions. This design eliminates the need for
block-based batching, supports real-time validation, and resolves conflicts by prioritizing higher-
weight transactions. Incorporating SHA-256 hashing for tamper resistance, Fernet symmetric
encryption for data security, and real-time CSV data exports for transparency, this architecture
ensures scalability, traceability, and efficiency, making it ideal for high-volume, dynamic
environments like procurement.
SHA-256 hashing and Fernet symmetric encryption are integral to securing the blockchain and
DAG architectures. SHA-256, a 256-bit cryptographic hash function, ensures data integrity by
generating unique, irreversible hashes for transactions, embedding each hash in subsequent ones
to create tamper-resistant links. Fernet symmetric encryption, leveraging AES-128 in CBC mode
with secure random initialization vectors, ensures data confidentiality by encrypting and
decrypting transaction records with a shared key. Together, these mechanisms protect against
tampering and unauthorized access, ensuring secure, transparent, and reliable supply chain
management.

2
Improving warehousing is vital for supply chain efficiency. Dynamic inventory prioritization
reduces retrieval times, while optimized storage maximizes space and cuts costs. Advanced
organizational strategies, such as demand-based storage allocation, automated sorting systems, and
real-time inventory tracking, enhance accuracy and streamline workflows. These enhancements
boost agility, minimize delays, and ensure resources align effectively with dynamic market
demands. The development of the Optimal Order Stacking (OOS) Algorithm, integrated with
LSTM-based demand forecasts, was guided by logical insights from the two referenced studies.
The first study, "An Efficient Technique for Arranging Various Commodities in a Warehouse,"
emphasized the importance of optimizing storage allocation and prioritizing high-demand
items[7]. These principles aligned with the OOS Algorithm’s design, which uses LSTM-generated
8-week demand forecasts to allocate inventory in Vertical Lift Modules (VLMs) based on
forecasted frequency and affinity. The second study, "An Overview of Warehouse Optimization,"
highlighted operational strategies like dynamic slotting and grouping, reinforcing the need for
efficient stacking mechanisms. The OOS Algorithm incorporates a heap-based priority queue to
dynamically manage excess inventory and maximize space utilization while maintaining forecast-
driven priorities[8]. Together, these studies validated the integration of predictive inputs with
algorithmic mechanisms to address challenges in prioritization, storage optimization, and
operational scalability. The developed Optimal Order Stacking (OOS) Algorithm optimizes
inventory allocation in Vertical Lift Modules (VLMs) using 8-week LSTM demand forecasts. It
prioritizes products based on frequency and affinity, employing a heap-based priority queue for
dynamic allocation. This ensures efficient space utilization, scalability, and streamlined warehouse
management.
Addressing key supply chain issues such as lack of vehicle capacity optimization, absence of
loading strategies, overlooked demand hierarchy, and integration gaps in predictive outputs
delivers substantial benefits. Optimizing vehicle capacity maximizes resource utilization, reduces
trips, and minimizes costs. Efficient loading strategies streamline operations, reduce labor, and
ensure safer transport. Prioritizing deliveries based on demand hierarchy enhances service
reliability and prevents stockouts in critical areas. Integrating predictive forecasts into scheduling
ensures proactive planning and effective resource allocation, resulting in a more adaptable, cost-
efficient, and service-oriented supply chain. The hybrid bin-packing and vehicle scheduling
algorithm was developed leveraging insights from "Bin Packing and Scheduling with Due Dates"
and "Research on Bin Packing Problem Based on Supply Chain Economic Efficiency." The first
study informed vehicle utilization strategies, prioritizing high-demand locations, allocating larger
vehicles first, and dynamically redistributing excess to smaller vehicles. It also introduced
scheduling constraints, such as limiting daily dispatches to 10 and aligning deliveries with demand
priorities[9]. The second study emphasized predictive analytics, enabling the integration of 8-week
LSTM-based demand forecasts for proactive planning, and iterative packing strategies for dynamic
allocation to handle fluctuating demand[10]. This approach ensures efficient vehicle loading,
adaptive scheduling, and compliance with capacity and delivery constraints. The developed
algorithm integrates vehicle scheduling and bin-packing to optimize delivery schedules and
vehicle utilization. Utilizing 8-week LSTM-based demand forecasts, it prioritizes high-demand
locations by allocating vehicles with assumed capacities of 20,000 and 10,000 units. A descending
capacity prioritization strategy maximizes utilization, dynamically redistributing excess demand
to smaller vehicles. To ensure balanced logistics, the algorithm limits daily dispatches to 10
vehicles and generates schedules based on demand hierarchies and temporal constraints. This
precise integration of predictive analytics, adaptive scheduling, and capacity optimization
enhances logistical efficiency and resource utilization.

3
The VRPTW model faces key shortcomings, including the lack of vehicle capacity considerations,
leading to underutilized resources and increased costs. It also fails to integrate predictive analytics,
limiting its ability to optimize routes based on demand forecasts. Additionally, the absence of
dynamic adjustments to fluctuating demand results in suboptimal delivery routes and delays.
Addressing these issues is essential for improving resource utilization, responsiveness, and
logistical efficiency. The design of the routing optimization algorithm is heavily influenced by
insights from " A New Approach to the Optimization of the CVRP through Genetic Algorithms "
and " Solving the Capacitated Vehicle Routing Problem with Environmental Criteria Based on
Real Estimations in Road Transportation: A Case Study " The first study emphasized addressing
capacity constraints and optimizing routes to meet delivery demands efficiently[11]. This informed
the integration of demand-proportional penalties, enabling the algorithm to prioritize high-demand
nodes and streamline delivery by excluding low-priority locations and nodes with multiple
delivery times. The second study provided the foundation for incorporating the Guided Local
Search (GLS) metaheuristic, ensuring dynamic route optimization by balancing vehicle capacity
and travel distance[12]. Together, these studies guided the development of a robust algorithm that
minimizes delivery costs while improving efficiency. The developed algorithm optimizes the
Capacitated Vehicle Routing Problem (CVRP) by integrating LSTM-based demand forecasts with
OR-Tools, ensuring efficient vehicle utilization within capacity and delivery constraints. The
model uses forecasted demands, vehicle capacities, and geospatial data to build a routing solution
based on a distance matrix and demand-proportional penalties for unmet demands. Node
prioritization is determined by demand thresholds, with low-priority nodes excluded dynamically.
Guided Local Search (GLS) refines the solution by minimizing route costs and balancing vehicle
loads. The final routes are visualized using Folium for enhanced logistical planning.
The sales data from Mint Paper Products Pvt. Ltd, spanning January 2015 to August 2024, formed
the backbone of this study, providing detailed monthly sales figures for face tissues, paper
serviettes, kitchen towels, and toilet rolls across multiple distribution centers. This dataset’s
dimensional standardization, designating face tissues as the unit size, paper serviettes as double
the unit size, and kitchen towels and toilet rolls as quadruple the unit size, was critical for
integration into advanced warehousing and bin packing algorithms. This standardization allowed
accurate spatial optimization, ensuring efficient inventory allocation and load balancing. Item wise
sales figures informed warehouse optimization by facilitating product specific storage strategies
and prioritizing high demand items for accessible placement, while location specific data enabled
tailored bin packing that maximized vehicle utilization and minimized inefficiencies. Additionally,
integrating item and location wise sales data with LSTM based forecasting models allowed for
precise demand prediction, enhancing decision making in both storage and distribution. In
procurement planning, the data was further utilized with LSTM based time series predictions that
combined historical sales trends with macroeconomic indicators such as the LKR to USD
exchange rate and line of credit interest rates established by the Central Bank of Sri Lanka. These
models analyzed fluctuations in exchange rates and credit costs to forecast procurement expenses
and determine the most cost effective purchasing dates. By correlating these economic factors with
supplier payment patterns, the algorithm identified optimal procurement windows, reducing costs
while ensuring inventory stability. With over nine years of granular sales records, the dataset’s
scale and specificity were particularly well suited to the fast moving consumer goods sector,
enabling a data driven approach to address warehousing, routing, and procurement challenges
effectively.

4
The developed system builds on the framework outlined in "An Innovative Machine Learning
Model for Supply Chain Management" by addressing its gaps and providing a synchronized
approach to supply chain operations. While the original study employs Conditional Generative
Adversarial Networks (CGANs) and predictive analytics, it lacks integration of vehicle capacity
optimization, dynamic inventory prioritization, and seamless coordination across supply chain
processes. The new system incorporates blockchain and Directed Acyclic Graph (DAG)
architectures to enhance traceability and transaction management, alongside the Optimal Order
Stacking (OOS) Algorithm, which leverages LSTM-based demand forecasts to efficiently allocate
inventory based on demand patterns. The system integrates the Bin Packing Algorithm (BPA) and
the Capacitated Vehicle Routing Problem (CVRP) model to improve logistics efficiency. The BPA
calculates the required number and configuration of vehicles by considering item volumes and
vehicle capacities, ensuring efficient space utilization. These configurations feed into the CVRP
model, which optimizes delivery routes by prioritizing high-demand nodes and balancing loads.
Vehicles with capacities of 20,000 and 10,000 units are dynamically allocated, with excess demand
redistributed to avoid inefficiencies. This synchronized approach connects procurement,
warehousing, and distribution, addressing the isolated processes in the original study and offering
a cohesive solution for supply chain management.
In conclusion, the developed system effectively addresses the limitations of the framework
presented in "An Innovative Machine Learning Model for Supply Chain Management" by
introducing a synchronized, scalable, and adaptable solution for modern supply chain
management. By integrating blockchain, Directed Acyclic Graph (DAG) architectures, the
Optimal Order Stacking (OOS) Algorithm, and the coordinated use of the Bin Packing Algorithm
(BPA) and Capacitated Vehicle Routing Problem (CVRP) model, the system enhances efficiency
across procurement, warehousing, and distribution. The synchronization between these
components ensures real-time adjustments to dynamic demand patterns, eliminates redundancies,
and optimizes key processes such as inventory allocation, vehicle utilization, and demand-driven
routing. The system’s computational efficiency can be significantly improved by leveraging
parallel processing techniques, such as CUDA. For instance, CUDA could be applied to parallelize
computationally intensive tasks like route optimization in the CVRP model. By dividing the
optimization process into smaller tasks executed concurrently on GPUs, runtime can be greatly
reduced, allowing for faster logistical planning. Similarly, CUDA can accelerate LSTM-based
demand forecasting by enabling rapid matrix computations over large historical datasets. These
improvements facilitate real-time decision-making even in complex or high-demand scenarios,
further enhancing the system’s operational efficiency.
The system’s adaptability allows it to be applied across industries such as healthcare, automotive,
electronics, pharmaceuticals, retail, and aerospace, each with unique supply chain needs. In
healthcare, it could optimize cold chain logistics for temperature-sensitive medicines, while in
automotive, it could streamline just-in-time supply chains for high-value components. For
electronics, the system can be tailored by adjusting the Optimal Order Stacking (OOS) Algorithm
to handle fragile, high-value components, incorporating product lifecycles and seasonal demand
patterns into LSTM-based forecasts, and ensuring compliance with regulations like
RoHS(Restriction of Hazardous Substances) and WEEE(Waste Electrical and Electronic
Equipment) using blockchain and DAG architectures. The CVRP model can also be adapted for
specialized transport needs, such as temperature and humidity control. These enhancements
highlight the system’s versatility in addressing industry-specific supply chain challenges.

5
References
[1] “Ferrero Sustainability News | Ferrero increases transparency of cocoa supply chain,” Ferrero,
2019. https://www.ferrero.com/int/en/news-stories/news/ferrero-increases-its-transparency-of-
cocoa-supply-chain (accessed Dec. 03, 2024).
[2] Amnesty International, “Forced Evictions at Industrial Cobalt and Copper Mines in the DRC,”
Amnesty International, Sep. 12, 2023. https://www.amnesty.org/en/latest/news/2023/09/drc-
cobalt-and-copper-mining-for-batteries-leading-to-human-rights-abuses/
[3] N. Afrin and A. Pathak, “Blockchain-Powered Security and Transparency in Supply Chain:
Exploring Traceability and Authenticity through Smart Contracts,” International journal of
computer applications, vol. 185, no. 49, pp. 5–15, Dec. 2023, doi:
https://doi.org/10.5120/ijca2023923318.
[4] R. Azzi, R. K. Chamoun, and M. Sokhn, “The power of a blockchain-based supply chain,”
Computers & Industrial Engineering, vol. 135, no. 1, pp. 582–592, Sep. 2019, doi:
https://doi.org/10.1016/j.cie.2019.06.042.
[5] M. S. Al-Rakhami and M. Al-Mashari, “ProChain: Provenance-aware Traceability Framework
for IoT-based Supply Chain Systems,” IEEE Access, pp. 1–1, 2021, doi:
https://doi.org/10.1109/access.2021.3135371.
[6] W. Yang, X. Dai, J. Xiao, and H. Jin, “LDV: A Lightweight DAG-Based Blockchain for
Vehicular Social Networks,” IEEE Transactions on Vehicular Technology, vol. 69, no. 6, pp. 5749–
5759, Jun. 2020, doi: https://doi.org/10.1109/tvt.2020.2963906.
[7] None Chhavi Gupta, “An Efficient Technique for Arranging Various Commodities in a
Warehouse,” Deleted Journal, vol. 31, no. 3s, pp. 265–276, Jun. 2024, doi:
https://doi.org/10.52783/cana.v31.764.
[8] J. Karasek, “An Overview of Warehouse Optimization,” International Journal of Advances in
Telecommunications, Electrotechnics, Signals and Systems, vol. 2, no. 3, Nov. 2013, doi:
https://doi.org/10.11601/ijates.v2i3.61.
[9] F. Koblasa and M. Vavrousek, "Bin packing and scheduling with due dates," in Proc. 36th Int.
Conf. Mathematical Methods in Economics, Jindřichův Hradec, Czech Republic, Sep. 2018.
[10] J. Zhu, “Research on Bin Packing Problem Based on Supply Chain Economic Efficiency,”
Advances in economics, business and management research/Advances in Economics, Business
and Management Research, pp. 343–354, Oct. 2023, doi: https://doi.org/10.2991/978-94-6463-
256-9_37.
[11] M. Frutos and F. Tohmé, “A New Approach to the Optimization of the CVRP through Genetic
Algorithms,” American Journal of Operations Research, vol. 02, no. 04, pp. 495–501, 2012, doi:
https://doi.org/10.4236/ajor.2012.24058.
[12] J. Faulin, A. Juan, F. Lera, and S. Grasman, “Solving the Capacitated Vehicle Routing Problem
with Environmental Criteria Based on Real Estimations in Road Transportation: A Case Study,”
Procedia - Social and Behavioral Sciences, vol. 20, pp. 323–334, 2011, doi:
https://doi.org/10.1016/j.sbspro.2011.08.038.