International Research Journal of Multidisciplinary Technovation

Optimal Fractional-Order PID Control for BLDC Motor Drives: A Robust and IOT-Enabled Approach

Megha Sharma, Shailly Sharma, Jayashri Vajpai, Venkataramanan V — Wed, 15 Apr 2026 00:00:00 +0000

Brushless DC motors have a very wide range of applications in industrial automation, however traditional PID controllers cannot provide stability with nonlinear and noisy operating conditions. To overcome this, a novel optimized Nelder–Mead algorithm is employed for an intelligent control structure using a Fractional Order PID controller, targeting the minimization of the Integral Time-weighted Absolute Error (ITAE). The proposed NM-FOPID framework is benchmarked against industrial standards, including Ziegler–Nichols (ZN) and Cohen–Coon (CC) methods, as well as metaheuristic benchmarks like PSO and GA. Experimental validation under different loading conditions (0.5–2.0 kg·m²) showed substantial performance improvement over traditional PID control. The FOPID controller reduced overshoot from 48.3% (traditional PID) to 8.24%, reduced settling time from 31.8 s to 5.9 s (≈ 81% improvement), and raised damping ratio to 1.73, leading to more robustness against disturbances. A White-noise test and frequency-domain analysis also verified high gain stability with a 25° phase margin. The proposed FOPID-based control realizes 70–80% improvement in transient performance and noise robustness, providing an optimal, Industry 4.0-compatible solution to smart BLDC motor control. Finally, the framework is implemented on a Raspberry Pi platform with Firebase integration, providing a scalable Industry 4.0 solution. The IoT layer achieves a measured jitter of ±1.2 ms and an 85 ms cloud latency, successfully decoupling high-speed local regulation from remote monitoring. These findings confirm that the NM-optimized FOPID provides a resilient, energy-efficient, and practical alternative for high-performance electric drive systems.

Endo-CNN: A Novel Deep Learning Model for Gastrointestinal Diseases

Esha Saxena, Suraiya Parveen, Mohd. Abdul Ahad, Meenakshi Yadav — Wed, 15 Apr 2026 00:00:00 +0000

Gastrointestinal (GI) diseases often represent the most frequent and common high-risk diseases. Wireless capsule endoscopy (WCE) has changed the landscape of diagnosing and treating patients. Endoscopists commonly utilize wireless capsule endoscopy to assess the majority of intestinal conditions, particularly with respect to polyps and ulcers. The use of WCE has shown a ten percent increase in Indian hospitals. Medical assessments are typically time-consuming and expensive, especially given the necessity to investigate directly from endoscopic videos. These confines are alleviated with the assistance of artificial intelligence and deep learning, which provide an efficient platform for instantaneous defect detection. The objective served by this examination is to assist endoscopic image classification work for clinical investigators. The paper proposed a deep-learning model named Endo-CNN based on convolutional neural network to classify endoscopic images according to the identified disease. The classes of images include polyps, ulcerative colitis, esophagitis and a healthy colon. Data augmentation occurs to reduce the imbalance of datasets and to evaluate the model performance that exceeds 48,000 images. The model achieves a positive accuracy rate with all the image classes. There are various aspects of an identified disease because of the variety of sizes, shapes and textures as well as colors. The paper also performs a comparative study of the designed model and against other pre-trained models. This paper can act as a baseline for many future solutions in the field of gastroenterology.

G-Transgan: Semantic Translation of Gujarati Texts using GAN-based Augmentation and Optimized Transformer Models in Low-Resource Settings

Mehulkumar Dalwadi, Abhishek Mehta — Mon, 20 Apr 2026 00:00:00 +0000

Gujarati is an Indo-Aryan language with more than 55 million speakers, making it an important language to consider in machine translation. It has limited parallel corpora, complex morphology, and no context preservation. The typical neural machine translation methods tend to fail in low-resource settings, resulting in syntactic errors and semantic drifts. To overcome these shortcomings, this paper presents Gujarati-Translation with Generative Adversarial Network (G-TransGAN), a new hybrid model that combines conditional Generative Adversarial Networks (cGANs), morphology-sensitive Sentence Piece tokenization, multilingual transformer embeddings (XLM-RoBERTa and Indic BERT), and optimization techniques such as Sharpness-Aware Minimization (SAM) and Low-Rank Adaptation (LoRA). The main goal is to maximize fluency, semantic retention, and domain flexibility in low-resource Gujarati-English translation. The workflow includes five steps: data augmentation, pre-processing and tokenization, contextual embedding, semantic translation, and optimization. The experimental findings indicate that G-TransGAN had better performance on various measures, including BLEU (38.4), METEOR (0.76), and TER (0.46). Such results support the model as able to produce high-quality, human-like translations and yet remain computationally feasible in low-resource real-world settings.