University of Technology Malaysia
Universiti Teknologi Malaysia Institutional RepositoryNot a member yet
70456 research outputs found
Sort by
Alzheimer’s disease classification using attention mechanism and global average pooling on a convolutional neural network
Get PDFThe robustness of Convolutional Neural Network (CNN) architecture as the innovative technology has led to the surge of research adoption for Alzheimer’s disease (AD) classification. CNN is replacing the conventional machine learning methods to assist and support experts in diagnosing AD. However, the performance of conventional CNN architecture in classifying AD class and Normal Control (NC) class is hindered by its behaviours that require a large-scale dataset. Nonetheless, the major hindrance in the AD domain is limited amount of dataset. Therefore, previous studies have adopted data-centric enhancement modules such as pre-processing techniques, data augmentation, and transfer learning strategies to improve the classification performance of CNN. Yet, these modules alone are still struggling to offer sound accuracy of classification of the disease due to CNN's overfitting issue and behaviour which is insensitivity to the local position information, also known as spatial invariance. A recent trend in this domain is the merge of an attention mechanism with CNN to enhance the classification performance. This is done by identifying and extracting the salient discriminative features of MRI images. However, the generalization ability is still hindered due to validity of one specific dataset found among many research works. This research then proposes a novel attention-based CNN model (AGap-CNN), that employs the global average pooling (GAP) to reduce the number of learning parameters to be used for classification by Softmax. The AGap-CNN combines the attention mechanism with the GAP layer for classification at the model header to enhance the classification performance of CNN and improve the generalization capability of the network. The AGap-CNN was validated on two benchmark datasets of OASIS and ADNI. Furthermore, in further analysing the network performance, the AGap-CNN was compared to the existing state-of-the-art methods. The proposed AGap-CNN model outperformed the existing state-of-the-art methods for the OASIS and ADNI datasets with 99.22% and 100% average validation accuracy, respectively. In other words, the proposed AGap-CNN model works with acceptable accuracy, sensitivity, and specificity in classifying AD class and NC class for both benchmark datasets of OASIS and ADNI dataset
Asic implementation of low latency montgomery modular exponentiation
Get PDFNowadays, electronic communication devices tend to design smaller in size, lighter in weight, lower in cost and higher performance. Individual may tend to use electronic communication devices when exchanging sensitive matters, such as personal details, contract documents, company secrets and specific passwords are sent to other parties. Since internet is one of the important key contacts and electronically communicates with billions of people, protection for the transmission of important messages over the internet is vital. Encryption plays a vital role for every user in ensuring security of communication within the organization. Hence the algorithms needed for safe communication. The motivation of this project is to protect digital data in computer confidentiality, as it is often stored on computer systems and distributed through the internet or other computer networks. Rivest-Shamir-Adleman algorithm is first introduced by Ron Rivest, Adi Shamir and Leonard Adelman in 1977, and it is known as one of the famous public key cryptography algorithms since it is an asymmetric cryptography. Besides, the theory behind RSA is relatively simple and easy for modification purpose as it relies on algorithm such as factorization and modular exponentiation. In this paper, the whole process and algorithm has been described for 256-bit key size. Due to the bit length of modulus, the work included different but suitable implementation, which is the basic, radix-4 and radix-16 implementations to reduce the speed of cipher-decipher process. Implementation on Verilog HDL using Vivado Design Suite software has been done. Enhancement on speed and delay is the main constraint of this project. According to the synthesis results, the radix-16 Montgomery Multiplier implemented in RSA cipher can be implemented with a nearly 60% reduction in encryption latency. However, radix implementation will involve more loop unrolling steps that resulted in a higher gate count. It is conceivable to absorb the increase in the gate count in the RSA cipher in return for performance as chip technology improves
Optimising energy performance of an Eco-Home using Building Information Modelling (BIM)
No full textIn a world where sustainability constantly manifests itself as the contemporaneous zeitgeist in practically every facade of our lives, it is imperative to understand the energy performances (EP) of buildings, both old and new, and explore innovative ways to optimise this cardinal aspect of building operation. In this light, we investigate the potential benefits of integrating Building Information Modelling (BIM) into EP analysis of built infrastructures. This research, in the form of a case study, has been designed to uncover equipments of high energy consumption in an Eco-Home, and subsequently to compare BIM-based simulations with the actual data measured on energy consumption. From the analysis, this study also proposes several recommendations on energy optimisation. For the purpose of EP analysis and simulation runs, Autodesk Green Building Studio (GBS) has been deployed, while 3D BIM Model of the Eco-Home was generated using Autodesk Revit. In situ energy audit revealed that the air-conditioner was the most energy-intensive equipment in the Eco-Home. Only a small variation of energy consumption was observed in actual and simulated data. Further analysis on design alternatives illustrated that the EP of the Eco-Home can be vastly improved by adopting a few measures, such as the installation of occupancy sensors to automate lighting, the integration of greywater reclamation system to reduce water consumption, and the addition of photovoltaic panels to increase renewable energy generation. However, from the same analysis, wind energy was found to be inviable due to its low level of energy potential. This paper concludes that integrating BIM and GBS into EP analysis not only improves the overall EP measurements, but also acts as an enabler for designers and building owners to compare design alternatives effectively. The advancement of BIM integration in this study offers an interesting proposition for the architectural, engineering, and construction industry especially when it comes to the enhancement of sustainability of a building
Improving the water quality of iron-containing ponds using fermented kitchen wastes
No full textTo elucidate the efficacy of eco-enzyme in mediating contaminated waters, an onsite investigation was conducted spanning over a 5-month duration on three ponds receiving iron-containing groundwater continuously. Eco-enzyme was applied at designated dosages and time. Water samples were collected fortnightly for in situ monitoring of water quality. Parameters determined included total suspended solids (TSS), pH, alkalinity, chemical oxygen demand (COD), biochemical oxygen demand (BOD5), ammoniacal-nitrogen (NH3), and dissolved oxygen. During the investigation, eco-enzyme application was halted for nearly a month owing to the enforcement of Movement Control Order to curb the rising Covid-19 cases. The results revealed that the application of eco-enzyme has improved the overall water quality of the three ponds. A 99.9% of TSS removal was achieved. For the other parameters measured, the concentration fluctuated over time and the water quality status of the pond interchanged between slightly polluted and clean. The elevation of the water quality was largely dependent on the frequency of the eco-enzyme application. This study, therefore, concludes that eco-enzyme possessed the capability to enhance the quality of contaminated water
Evaluation of sneak path current in self-rectifying memristor crossbar
Get PDFMemristor is a non-volatile memory nanodevice which requires lower energy to operate. Besides, memristor is fast yet stable, thus making it attractive in the computing field and competitive as alternative candidate for the integration of high-density memory. While acting as a memory cell in a crossbar array circuit, there is limitation related to undesired sneak path current. Utilizing a self-rectifying memristor is reported to minimize the effect of the sneak path issue. Since there is lack of resources on research of implementing self-rectifying memristor in crossbar array structure, it is the focus of this project. The objective of this project is to model and simulate self-rectifying memristor with different rectification ratio and non-linearity. Their effect on the performance of memristor in crossbar array structure is then analyzed. Several types of self-rectifying memristor are reviewed and their compact SPICE models are generated. The memristor SPICE model is used to build a crossbar array circuit and run simulation through LTSPICE software. Moreover, comparison between Knowm memristor and self-rectifying memristor is done by observing their current-voltage relationship. The reduction of sneak path current by using different value of saturation current in self-rectifying memristor is evaluated and the effect of rectification ratio and non-linearity is being studied. Based on the simulation result, low saturation current helps to ensure large rectification ratio and non-linearity to lower sneak path current in the circuit. In short, self-rectifying memristor is able to effectively suppress sneak path current in crossbar array
Fresnel lens defect classification using deep learning technique
No full textPlastic injection molded Fresnel lens is one of the important components for illumination in smart devices. To perform inspection on this type of optical component is challenging for machine vision due to the presence of groove pattern and texture. This paper discusses the limitation of classical image analysis for defect inspection and proposes a Deep Convolutional Neural Network (CNN) with Transfer Learning for defect classification. This paper also presents a Hybrid CycleGAN and geometric augmentation to expand image dataset for model training
Graphene-polyvinyl alcohol polymer based saturable absorption at 2000 nm region
Get PDFA graphene-polyvinyl alcohol (PVA) composite saturable absorption is demonstrated at 2000 nm region. Graphene suspension is produced using low-cost electrochemical exfoliation process. The suspension is mixed with PVA host polymer in 1:1 ratio and left evaporated at room temperature which finally produced graphene-PVA thin film. Thulium doped fiber (TDF) gain medium has been shown to produce a stable Q-switched pulse with a highest repetition rate of 54 kHz, a short pulse duration of 2.89 µs, a maximum peak power of 16 mW, and an estimated maximum pulse energy of 49 nJ. Apparently, at 2000 nm region, superior performances of graphene-PVA composite have been recorded which was largely contributed by meticulous composite preparation and homogenous mixture with PVA host
An efficient march (5n) FSM-based Memory Built-in Self-Test (MBIST) architecture with diagnosis capabilities
Get PDFIn deep submicron Systems-on-Chip, embedded memories are consuming a growing part of the die area. The manufacturing test of embedded memory is a critical stage in the SoC production process that screens out faulty chips and speeds up the volume production of new manufacturing technology. Memory Build-In Self-Test or MBIST is a standard mechanism to test the memory arrays and potentially detect all of the faults that may be present inside memory cells using an effective collection of algorithms. However, a massive number of memory cells wrapped by BIST logic can result in substantial overhead in wiring and gate area, and also a detrimental influence on memory performance. Therefore, new MBIST designs for advanced SoCs that address the challenges must be explored to reduce the overall cost of manufacturing tests. It is important to choose the appropriate level of algorithmic coverage and diagnosis for a range of array sizes. The March 5n algorithm proven the alternative form of March-based algorithm with better test length has achieved shorter test time than conventional MATS++ algorithms without penalizing the fault coverage. This memory testing algorithm and architecture suit the needs for fast array testing to get the products to market in the quickest fashion. However, the previous work is extendable for inversion coupling fault detection and repair support. Therefore, the March 5n architecture is utilized as the foundation in this project. An improved March 5n architecture is proposed to extend its properties in terms of fault coverage and diagnosis capabilities to allow memory failure analysis. Block of March algorithms, an address generator, data generator, diagnosis module, and redundancy logic are the components of the targeted BIST architecture. Extensive circuitry from the previous architecture will be implemented to achieve the goals. The additional logic will accumulate the fault information and its corresponding diagnosis results will report during the memory testing. Synopsys Electronic Design Automation tools (VCS, Design Compiler and Verdi) are utilized in synthesising and evaluating the performance in terms of speed, area, power and fault coverage. Several reports and waveforms are generated and simulated for evaluation. The outcome of this project has demonstrated that adding more logic can enhance the capability for diagnosis and enable redundant programming to replace the defective cell. Besides, the inversion coupling fault coverage using the March 5n is verified to be functioning as intended. Speed up of the redundant memory space allocation in a repair mechanism is achieved with the proposed architecture due to the ability to keep track of each failure signature of memory when tested. In comparison to earlier work, the improved architecture has generally enhanced maximum clock speeds by almost 8% and decreased power dissipation by about 6%. However, higher speed and functionality are obtained at the cost of 4% of the area overhead
Performance evaluation of single-stage differential amplifier based on carbon nanotube
Get PDFThe demand for power sensitive CMOS designs has grown significantly due to the fast growth of battery-operated portable applications. Design of low-power and high-performance submicron and deep submicron CMOS circuits has become a big challenge in nanoelectronics industries due to short-channel effect that occurs after scaling towards nanoscale devices. Silicon-based power transistor devices has low power consumption which allows more components per chip surface area. But silicon-based short-channel devices has generated DIBL effect, hot carriers’ effect and surface scattering, results in device performance degradation. To overcome these unwanted effects, carbon nanotube-based devices has shown the potential to replace silicon-based devices by sustaining the requirements of a high-speed nanodimensional devices because it has similar device operation with CMOS and produces lower leakage power than silicon-based devices. Differential amplifier circuit topology is applied in this research because it is a very useful operational amplifier circuit to examine the performance differences between carbon nanotube and conventional silicon when they are used as channel material by evaluating Common-Mode Rejection Ratio (CMRR) of differential amplifier. The objective of this research is to study the performance of Carbon Nanotube based differential amplifier based on CMRR and to compare the performance of Carbon Nanotube based differential amplifier with the silicon based differential amplifier. HSPICE tool is used in this research to simulate the differential amplifier circuit with current mirrors active load configuration to maintain the voltage gain for single-ended output, which is built using netlists of SPICE CNFET model and PTM model, respectively. From the research findings, the highest CMRR of CNFET-based differential amplifier with constant input DC offset voltages in differential mode and common mode is 72.68 dB. When input DC offset voltages in differential mode and common mode decreases, CNFET-based differential amplifier has achieved CMRR of 92.16 dB, which increases by 26.8% compared to that of constant input DC offset voltages. The CMRR of MOSFET-based differential amplifier (21.83 dB) is smaller than the CMRR of CNFET-based differential amplifier (132.02 dB), with a difference of 110.19 dB or 143.2%
Synergistic application of polypropylene and silica nanoparticle modified by (3-Aminopropyl) triethoxysilane for cuttings transport
Get PDFAchieving the desired efficiency in cuttings transport has always been a challenge during drilling situations due to complex wellbore trajectories and drilling hydraulics. These issues require suitable additives to formulate the drilling muds. Hence, nanocomposite (NC) particles formed from silica and polypropylene was modified by (3-Aminopropyl) triethoxysilane (termed PP-SiO2 NC-NH2) and the impact of the concentrations of the PP-SiO2 NC-NH2 between 0.4 and 1.2 ppb on the cuttings transport efficiency (CTE) of complex water-based mud (WBM) system was assessed. Mud system containing 0.5 ppb PP-SiO2 NC-NH2 on CTE was measured and compared with that of partially hydrolyzed polyacrylamide (PHPA) using sandstone grains of diameters between 0.50 and 3.20 mm at hole angles 0, 30, 45, 60, and 90° with a 42 L/min pump rate. The zeta potential (?-potential), particle size distribution, morphology, and temperature resistance of the NC were examined. The ?-potential data revealed the stability and the long-term stabilization of the modified NC in drilling muds. By adding the concentration of the modified NC into the WBM system, the thinning characteristics and consistency factors were enhanced as the NC concentration increases. Furthermore, the rheological test program reveals that the NC concentrations and 0.5 ppb PHPA increased the shear stress of the WBM with an increase in the shear rate. However, the mud system of PHPA exhibited a larger viscosity, while the NC samples showed a flat viscosity-profile. Also, the NC demonstrated better cuttings recovery than the PHPA product at 0.5 ppb concentration, especially at the worst hole angle of 45°. The cuttings recovery from a minimum to a maximum was found to occur in this order, 45°, 60°, 30°, 90°, and 0°. In contrast with the smallest and intermediate sand grains, the largest grains have complex and less fluid transport. With a pipe orbital motion of 150 rpm and a pump rate of 42 L/min, the CTE of the muds showed higher improvement. Overall, the mud properties of the modified NC exhibited a strong enhancing impact on the cuttings carrying capacity of the base mud than the PHPA