FATMLPGS: Design of a Fault-Aware Trust Establishment Model for Low-Power IoT deployments via Generic lightweight Sidechains

: Networks connected to the Internet of Things (IoT) are constantly under attack from a wide range of adversaries, such as Distributed Denial of Service (DoS), Grey Hole (GH), Sybil, Masquerading, Spoofing, Man in the Middle, etc. The network may be under attack from adversaries on the inside or the outside, which lowers the QoS performance (Quality of Service) in terms of end-to-end delay, throughput, energy use, and other metrics. Researchers have proposed a number of security & privacy preservation mechanisms to counter these attacks, and each one differs in terms of computational complexity and security levels. The immutability, traceability, transparency, and distributed nature of blockchain-based models make them among these mechanisms highly effective in terms of security performance. These models are not suitable for large-scale IoT Network deployments because the QoS performance of these models is highly reliant on the length of the blockchain. Researchers have suggested sidechaining techniques as a way to get around this restriction because they help to increase QoS performance while still being highly secure. Machine Learning Methods are needed to determine when to split or merge the chains when designing sidechaining models because they are so complex. For low-power IoT networks, these models are therefore useless. This text suggests a lightweight Multiple Objective Grey Wolf Optimization (MGWO) Model that aids in initial route establishment by choosing high-trust nodes, reducing power consumption during sidechaining and incorporating fault-aware trust establishment. The number of pieces the blockchain must be divided into in order to maintain high QoS performance is also determined by the MGWO Model. In order to help the MGWO Model incrementally train to detect network faults, the model also includes a lightweight Q-Learning layer. An Intrinsic Genetic Algorithm (IGA) that is stochastically modelled and activated by the MGWO Model based on the network's progressing QoS performance in various communication scenarios controls the fault identification process. The proposed model can mitigate attacks such as Sybil, Masquerading, Grey Hole, Distributed Denial of Service (DDoS), and Man in the Middle (MITM) thanks to integration of Q-Learning with MGWO & IGA. The proposed model was found to be able to maintain high QoS even when under attack, which helps to increase the efficiency of its real-time deployment. When compared to different state-of-the-art techniques, the proposed model was found to have 15.9% better energy efficiency, 10.6% better throughput, 8.3% faster communication speed, and 0.8% higher packet delivery performance under different scenarios.


Introduction
The design of encryption, hashing, privacy preservation, route establishment, attack mitigation, fault detection, and QoS aware modelling techniques are required for trust establishment with security in IoT networks. Researchers have suggested a wide range of trust-based methods for designing these models, and each of them has unique nuances, benefits, limitations, and future research potential. Figure  1 shows an example of a blockchain-based trust establishment model where data security is enforced using single chained Byzantine Fault Tolerance (BFT) based blockchains. Where, represents total number of blocks in the chain, ( ), ( ), ( ℎ ), ( ), and ( ℎ) represent delays needed to read a block from the chain, write a block to the chain, check the block's hash value, verify uniqueness of the block, and hash the block before adding to the blockchain. In such cases, delay needed to add a block increase exponentially w.r.t. length of the blockchain, which limits its real-time deployment capabilities.
To overcome this issue, various sidechaining models are proposed by researchers, and a survey of such models [2,3,4] is discussed in the next section of this text. This survey also includes study of different fault tolerant & trust establishment models, which would assist readers to identify their working characteristics. Based on this review, it was observed that QoS of these models is highly dependent on complexity of blockchain mining, delay needed to establish trust, and complexity of fault detection models. Thus, most of these models have lower QoS performance, which limits their real-time scalability characteristics. To overcome these limitations, section 3 discusses design of a fault-aware trust establishment model for low-power IoT deployments via generic lightweight sidechains. The model uses a combination of Multiple Objective Grey Wolf Optimization (MGWO), Q-Learning, and Intrinsic Genetic Algorithm (IGA) in order to improve its QoS & security performance. The model was evaluated in section 3, w.r.t. various state-of-the-art models, in terms of various QoS metrics. Based on this evaluation, readers will be able to identify nuances, advantages & limitations of the proposed model w.r.t. standard trust-establishment methodologies. This paper finishes with some insightful observations regarding the suggested model and offers a number of suggestions for enhancing it.

Review of existing models
Researchers have proposed a wide range of trust formation models that support numerous scale network deployments with little complexity and high security performance levels. For instance, work in [5,6] proposes use of Optimized User-Friendly Transaction Time Management and Object Name Service (ONS), which helps with performing fundamental blockchain deployments for various network types. However, these models' usability is constrained by the fact that they are not scaleable. The work in [7] proposes the design of the Dynamic Ordered Node Selection (DONS) model, which helps with low complexity and high density blockchain-based access control under various network scenarios. Similar models are covered in [8,9,10], which advocate the use of Edge Computing, Bitcoin-NG, and Byzantine fault tolerance (BFT) based consensus models to support high scalability and low complexity trust establishment with delay minimization for various network types. Extensions to these models are discussed in [11,12,13], which recommend using Authentication with Key Management, Collaborative Internet-of-Things, and Space-Structured Greedy Heaviest-Observed Subtree (S2GHOST), thereby minimizing overheads while deploying various blockchain network types. Work in [14,15,16] proposes the use of Adaptable PBFT, lattice constructions, and BeeKeeper Model, which uses low complexity blockchain methods for integrating into highly scalable networks, to further enhance these models.
Models that further develop blockchains for Distributed Charging-Record Management [17], service-oriented architecture (SoA) [18], Deep Reinforcement Learning (DRL) [19], and Federated Learning (FL) [20] also make use of a variety of fault tolerance mechanisms that can be implemented for high-efficiency network applications. These models aim to reconfigure networks at runtime, which restricts their performance for real-time & high-speed application use cases. To address these problems, work in [21,22,23] suggests using Time Protection Schemes, short signatures with hash functions, and lightweight blockchains that enable adaptive reconfigurations without overwhelming the network with beacon signals. These models are further expanded through the use of application-specific blockchain deployment tools like Service-Oriented Permissioned Blockchain [24], Inhomogeneous Poisson Point Process (IP3) [25], Single Point of Failure (SPOF) [26], Federated Learning (FL) [27], and Transactive Energy Models (TEMs) [28]. These models improve network energy efficiency for various scenarios and reduce mining complexity by pre-empting search spaces. Extensions to these models are discussed in [29,30,31], which recommend using faulty probability determined (FPD) with faulty number determined (FND), Sybil-proof wireless network coordinate-based Byzantine consensus (SENATE), and Mobility Aware Blockchain-Enabled Offloading Schemes (MABOS) to help deploy large-scale networks that can be used for low-complexity, high-speed deployments. However, the QoS performance of these models is strongly influenced by the size of the blockchain, making it impossible to scale them for more extensive deployments. Even the creation and maintenance of sidechaining models requires complex machine learning techniques to make split or merge decisions for various chain types. They cannot be used for low-power applications because they use complex models. The design of a novel, lightweight Multiple Objective Grey Wolf Optimization (MGWO) process is covered in the following section in order to lower this power consumption during sidechaining and incorporate fault-aware trust establishment. The model's performance was validated for various real-time application scenarios by evaluating it under various QoS, fault, and attack scenarios and comparing it to various cutting-edge models. To effectively describe flow of the proposed model, it is segregated into different sub modules, and each of these modules are discussed in separate sub sections of this text. Based on this discussion, researchers will be able to implement these models for their applicationspecific deployment use cases.

Design of the GWO Model for Sidechain & Fault Tolerant Route Identification
The GWO Model is used to decide sidechain management operations, and to establish fault tolerant routes based on temporal node performance. The model initially collects nodelevel & network-level information that includes temporal communication delay, temporal throughput, packet delivery performance, and energy consumption per communication. This information is processed by a GWO Model, which works via the following process, Wolves', which indicates that these wolves must be modified during initial iterations • For each GWO iteration between 1 to , perform the following tasks, o Scan each of the wolves between 1 to , and analyze them via the following processes, ▪ If the wolf is marked as 'Alpha', 'Beta', or 'Delta', then go to the next wolf in sequence ▪ Else, generate wolf configuration via the following process, • Add dummy blocks to the current blockchain, by sending requests from requesting list of sources ( ) to requesting list of stochastic destination ( ) nodes.
• Identify paths with stochastic hops, which are evaluated via equation 2,

=
(1, ℎ ) … (2) Where, ℎ represents maximum allowable hops for the given network configuration, and can be setup by network designers.
• To communicate data between these node pairs, identify nodes, that satisfy equation 3, Where, ( , ) & ( , ) represents distance between source & current node, and distance between destination & current node, while represents distance between source & destination node pairs.
• Based on this selection, communicate data packets between nodes, and add blocks to the blockchain for each communication sequence.
• For each block addition, evaluate single wolf fitness via equation 4, In the table represents best fitness levels for final iteration, while ( ) represents best fitness levels for current source & destination pairs. Based on this evaluation, select the path with highest ( ) levels for routing the data, while evaluate Sidechain Split Function ( ) via equation 7 to estimate if the sidechain is needed to be split or merged with other chains.

= ( ) … (7)
After this evaluation, the current sidechain is needed to be split if > 1 , which indicates that current sidechain performance is low, thus it is needed to be split into equal parts. But, if < 1 2 * , then the blockchain must be merged with largest length chain, and a new sidechain must be created for future communications. Otherwise, there is no need to modify sidechain configurations. Due to these simplistic decisions, the model is capable of deploying low-energy, and high-efficiency sidechains with better trustlevel routes. This model is cascaded with an IGA technique, which decides if model retraining is needed, thereby reducing the effort needed to reconfigure existing sidechains. Design of this model is described in the next section of this text.

Design process for the IGA Model to integrate QoS awareness retraining decisions
The GWO Model recommends split & merge decisions, and trust-based routes, but it cannot be re-evaluated for every communication sequence. Because re-evaluating the model will require higher energy consumption, which will reduce its deployment capabilities. Thus, to avoid frequent reconfiguration of routes, an IGA Model is deployed, that continuously scans communication requests, and maps them with existing GWO evaluated paths. This model is activated for every new communication request, and works via the following process, • Initially setup the following IGA parameters, Where, ( ) represents selected path, ( ) & ( , ) represents number of nodes possible between current source & destination, and current node ID used for correlation matching process.
Where, ( ) represents the fitness evaluated by the IGA Model, and is used to evaluate reconfiguration decisions. This process is performed for communication requests, and for each request a counter is incremented if > , which assists in confirmation of reconfiguration decisions. At the end of communications, if the value of > * , then the model doesn't need reconfiguration, else the GWO Model is re-evaluated and table 1 is appended with new route combinations. This process assists in reducing energy requirements by reducing unwanted model evaluations. To further enhance this performance, next section discusses design of a Q-Leaning model, which assists in auto updating table 1 for better correlation-based matching performance under real-time communication requests.

Design of the Q-Learning model with incremental learning process
The IGA Model is useful for improving QoS performance via reducing number of GWO reconfiguration processes. This model is further optimized via use of a Q-Learning model, that assists in estimation of reward functions for every IGA iteration sets. The Q-Learning model identifies highest fitness solutions for each IGA iteration, and estimates a reward function via equation 12 as follows, If the value of reward function for this iteration satisfies equation 13, then solution of current IGA iteration is added to table 1, else it is discarded from list of solutions. The accepted solutions are use to incrementally update routing configurations, which reduces complexity during selection process. Due to which, energy efficiency of the model is improved. This efficiency along with evaluation of other QoS & security related parameters under different attack types is discussed in the next section of this text.

Results & Comparison
The  [13], DRL [19], and SEN ATE [30]. To evaluate this performance, standard network configuration parameters were used, which can be observed from  The results of this evaluation on a smaller network size show that the proposed model showcases 10.5% lower energy consumption than S2GH OST [13], 16.3% lower energy consumption than DRL [19], and 12.5% lower energy consumption than SEN ATE [30] under different communications. These results are based on the fact that the proposed model consumes 10.5% less energy than S2GH OST [13]. This is because of the integration of residual energy as well as temporal energy consumption levels throughout the route selection and mining process, which enables better scalability performance to be maintained. On the basis of a comparable approach, the following can be noticed about the energy consumption performance of 500 nodes from  [19], and SEN ATE [30]. This is because of the integration of residual energy as well as temporal energy consumption levels throughout the route selection and mining process, which enables better scalability performance to be maintained. On the basis of a comparable approach, the following may be seen about the energy consumption performance of 1000 nodes, which can be found in The results of this evaluation on a large network size show that the proposed model exhibits 14.5% lower energy consumption than S2GH OST [13], 18.3% lower energy consumption than DRL [19], and 16.5% lower energy consumption than SEN ATE [30] under various communications. These results can be seen by observing that the proposed model consumes 14.5% less energy than S2GH OST [13]. This is because of the integration of residual energy as well as temporal energy consumption levels throughout the route selection and mining process, which enables better scalability performance to be maintained. The following may be noted for the average throughput performance for 100, 500, and 1000 nodes by consulting  On the basis of this evaluation on small, medium, and large network sizes, it can be seen that the proposed model demonstrates 19.5% better throughput than S2GH OST [13], 19.8% better throughput than DRL [19], and 8.5% better throughput than SEN ATE [30] under various communications. These results are achieved by comparing the proposed model to S2GH OST [13], DRL [19], and SEN ATE [30]. This is because the temporal throughput levels have been included into the route selection and mining process, which has allowed for improved scalability performance to be maintained. The following is an analysis of the average Packet Delivery Ratio (PDR) performance for 100, 500, and 1000 nodes that is based on a similar method and can be found in  On the basis of this evaluation on small, medium, and large network sizes, it can be seen that the proposed model exhibits 6.5% better PDR than S2GH OST [13], 8.5% better PDR than DRL [19], and 5.9% better PDR than SEN ATE [30] under various communications. These results can be seen when comparing the proposed model to S2GH OST [13], DRL [19], and SEN ATE [30], respectively. This is because temporal PDR levels are included into the route selection and mining process. As a result, better levels of scalability performance are maintained. These assessments are expanded upon for a variety of different kinds of assaults and are summarized in the next portion of this article under real-time attack scenarios.

Evaluation & Comparison of security levels
For the purpose of implementing a greater level of security, the model makes use of sidechains. This enables the network to defend itself against a variety of attack types, such as Sybil, DDoS, MITM, Grey Hole, and so on. In this part, the number of attacker nodes was changed from 0.5 percent to 10 percent, and the quality-of-service performance was analyzed for a variety of attack methods. This performance was compared with S2GH OST [13], DRL [19], and SEN ATE [30], which facilitates in detection of performance increase when compared with other models. [13], [19], and [30] On the basis of this method, the delay in communication during a Sybil assault relative to looking at  Table 11. Average end-to-end delay for security models (WH Attack) The results of this assessment reveal that the suggested model has minimal delay levels in comparison to the S2GH OST [13], DRL [19], and SEN ATE [30] models. This can be seen by looking at the results of the evaluation. However, in the case of our deployment, the delay demonstrates a continuous trend, which makes it relevant for attack scenarios. This delay should rise during assaults; however, this does not happen. Similar findings were made for the MITM and DDoS attack types based on the information in Based on this assessment, it can be shown that the suggested model has a reduced energy consumption when compared with S2GH OST [13], DRL [19], and SEN ATE [30] models even when subjected to Sybil, DDoS, and MITM attack types. This is something that can be noticed. This is as a result of the use of blockchain as well as sidechain models during the development of the underlying network for various use cases. Similar findings were observed with regard to throughput under Sybil, DDoS, and MITM attack types, as shown in

Comparative performance under different fault levels
Additionally, temporal characteristics are accounted for inside the model during the discovery of various route configurations. Because of this, the model is exceptionally adept at minimizing errors across a wide variety of network configurations. In this part, we analyze the performance of the model when subjected to a variety of node failures and compare it to the results obtained by the S2GH OST [13], DRL [19], and SEN ATE [30]  by 18.5% in terms of PDR performance. This enables it to be used for a broad range of realtime applications, which expands its potential uses. As a result, the fundamental FATMLPGS model enhances the quality of service (QoS) performance, attack resistance, and fault mitigation across a variety of node and network configurations. As a result of these improvements, it is applicable to a broad range of different real-time network deployment use cases.

Conclusions and Future Scope
The suggested model first utilizes a GWO based dense learning method to predict node configurations & sidechaining configurations for QoS & security aware performance. The model makes use of a lightweight genetic algorithm (GA) model, which supports in estimate of reconfiguration choices using correlation-based matching approaches. This is done so that the complexity of training may be reduced. This approach also helps in estimating fault-free pathways throughout the routing process, which is one of the many reasons why it is very beneficial for deploying real-time network infrastructure. The model will eventually include a Q-Learning strategy, which will help improve its performance by gradually upgrading its route and routing settings. This will be accomplished. When compared with other models, it was found that the suggested model had superior quality of service performance, stronger attack mitigation performance, and improved fault tolerance. It was discovered that the proposed model exhibited 5% less delay than S2GH OST [13], 14.5% less delay than DRL [19], and 18.4% less delay than SEN ATE [30]. Additionally, it exhibited 10.5% less energy consumption than S2GH OST [13], 16.3% less energy consumption than DRL [19], and 12.5% less energy consumption than SEN ATE [30]. Additionally, it exhibited 19.5% more throughput than S2GH OST [13], 19.8% more throughput This is because temporal PDR levels are included into the route selection and mining process. As a result, better levels of scalability performance are maintained. It can also be noticed that suggested model highlights reduced latency, low energy consumption, greater throughput, and higher PDR levels when compared with S2GH OST [13], DRL [19], and SEN ATE [30]  performance, attack resistance, and fault mitigation across a variety of node and network configurations. As a result of these improvements, it is applicable to a broad range of different real-time network deployment situations. In the future, researchers will be able to improve the performance of this model by integrating reinforcement learning models with a modest level of complexity. They may also enhance model's performance by integration of Convolutional Neural Networks (CNN), LSTM & GRU Models, which will aid in pre-emptive actions for higher performance under varied attack & fault levels.

Ethics approval and consent to participate
While preparing this manuscript, it does not report on or involve the use of any animal or human data or tissue Thus it is not applicable in this section.

7.Consent for publication
The manuscript does not contain data from any individual person. Hence it is not applicable in this Section.

8.Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

9.Competing interests
The authors declare that they have no competing interests.

Funding
There is no funding Support from the third party.

Authors' contributions
While preparing this manuscript Ashutosh Kumar Choudhary has done the simulation and analysis work while Surendra Rahamatkar has done result and analysis part.