AI Driven Crop Guardian For Farmers

AI Driven Crop Guardian For Farmers

Dr. Sandhya Deshpande

Electronics and Telecommunication Engineering

K.J Somaiya Institute Of Technology Mumbai, India

Ms. Rishika Shetty

Electronics and Telecommunication Engineering

K.J Somaiya Institute Of Technology Mumbai ,India

Ms. Krisal Naidu

Electronics and Telecommunication Engineering

K.J Somaiya Institute Of Technology Mumbai,India

Ms. Shrushti Surwade

Electronics and Telecommunication Engineering

K.J Somaiya Institute Of Technology

Mumbai ,India

AbstractAgriculture still plays a very large role in economy and food security issues, yet farmers also struggle with planting the right crops, depleting soil of nutrients, and weather which changes without notice. In the traditional approach of experience based farming which we see today, we see that it plays a large role in farm decisions which in turn leads to sometimes poor performance and profit. In this paper we present a smart crop recommendation and soil health prediction system that uses soil nutrient analysis and machine learning to better the farmers lot. We put into play here a system which looks at key soil elements of Nitrogen (N), Phosphorus (P) and Potassium (K) as well as also looking at environment and crop production variables. Soil quality is assessed and optimal crops and fertilizers are recommended using a supervised learning model involving Linear Regression and Random Forest classification. The system's data-driven recommendations enhance crop productivity, minimize manual labor, and support sustainable farming. The aim of the solution is to assist farmers in prudent decision-making, relieve the anxiety associated with uncertain crop performance, improve their yield, and reduce their expenses.

1. Introduction

   India's growth as a society and an economy can be largely attributed to the nation's strong agricultural sector. However, as the problems of modern farming continue to mount, the nation has suffered due to poor crop selection, declining soil fertility, unpredictable weather, and an undereducated agricultural workforce. Many farmers make decisions based on their own experience rather than scientific farming, leading to reduced crop yields and contributing to economic instability. Additionally, the overuse of chemical fertilizers and irrigation has negatively impacted soil health and the environment.

   Tech moves fast and loads of agriculture data exist; data analytics plus machine learning? Huge chance for sharper farming calls. By analyzing soil properties and environmental conditions, it is possible to predict crop performance and recommend the most profitable and suitable crops for cultivation. Soil nutrients such as Nitrogen (N), Phosphorus (P), and Potassium (K), along with parameters like rainfall, temperature, and market demand, play a crucial role in determining crop productivity

   AI Driven Crop Guardian, the proposed system, machine learning algorithms such as Linear Regression, XGBoost, Resnet and Random Forest are employed to inspect the nutrient level in soil and to decide how healthy the crops can be grown there. The system informs you of cheapest, most productive crop for your soil type, and if necessary provides fertilizer advice to improve fertility of said soil. The objective of the data-based crop advisory model would be to assist farmers get accurate, scientific information, do away with guess work and ensure higher crops output and more adoption of environment-friendly farming techniques.

2. Literature Survey

   Recent advances in precision agriculture have emphasized the importance of integrating soil information, weather data and machine learning for better crop yield. Wu et al. proposed a geo-object-based method for mapping of SOM based on high-resolution remote sensing and ensemble leaning approaches such as Random Forest (RF) and extreme gradient boosting algorithm (XGBoost), in order to take advantage of higher accuracy compared with grid-based maps for making predictions of soil properties more accurate [1]. Their study demonstrated that contemporary machine learning is particularly useful in dealing with the non-linear relationships between soil variables and environmental factors.

   Smart soil monitoring leveraging the Internet of Things? Folks built those to automate measuring things like moisture temperature and pH. Ananthi and team? We built a way for planting soil to tell farmers when it needs water using signals sent without wires. Suggesting crops from afar is possible now, thus less field work, right? These systems help with sensing at the field level, but they don't have the ability to predict which crops to choose. Patil et al. An intelligent prediction model based on neural networks was created to estimate cyclical populations of cotton field pests, Thrips tabaci. This model showed high accuracy when tested on surveillance datasets [3]. Predicting pests well helps with crop care yet needs integrating into environmental or soil-based tips so decisions improve throughout.

   Kulkarni et al. proposed an ensemble based model combining Random Forest, Naïve Bayes, and Linear SVM classifiers to predict the best crops that can be grown in specific soil and climate conditions. The model achieved very high accuracy through majority voting [4]. Asolkar et al. studied the use of GSM communication for monitoring crops in greenhouses. This is important for high-value crops, as it helps to control and stabilize the humidity, temperature, and irrigation [5]. Yield forecasting studies include research in both statistics and machine learning. Shakoor et al. reported on the use of supervised learning in sets of past agricultural production data to determine which are the most cost effective crops for certain regions [6]. Also in contrast Altlar et al. looked at the use of regression based models for prediction of wheat yield in Turkey which also brought to light the importance of model selection which takes into account climate variation and annual fluctuation [7]. Remote sensing plays a key role in identifying what crops are and how much they will yield. Miao et al. used multi temporal Landsat info along with Support Vector Machine (SVM) for crop classification in very broken up farmlands which they did very well when they used optimal temporal bands. Little et al. looked at MODIS NDVI time series to do farm level prediction of cotton yields which they did with great success [9].

3. AGRICULTURAL RELEVANCE

   In India agriculture still plays a key role in food security and rural economy. To sustain agricultural growth we must use our soil resources well and choose which crops to plant carefully. But farmers also do not always have access to soil analysis which in turn leads to poor crop choices, nutrient imbalances and low yields. We have put together a system which puts soil nutrient testing with machine learning and Gen-Ai based crop recommendation which in turn helps farmers to base their decisions on data.

   This work is relevant in areas which are very much at the mercy of soil fertility and annual rain fall. For example in the Bodvad taluka and in villages like Engaon which are in the Jalgaon district of Maharashtra we see semi arid agricultural zones which see great variation in soil nutrients and large scale effect of weather on crop output. The study looked at key soil elements which are Nitrogen (N), Phosphorus (P), and Potassium (K) and out of these based the best crop type as well as the fertilizer needs. This allows farmers to optimize the use oftheir land, minimize fertilizer overuse, and sustain healthy soil for longer. The system simplifies crop yield forecasting by providing more precise, localized recommendations for particular crops. Consequently, farmers no longer have to depend solely on their own past experiences to make decisions. Technology-enabled support like this boosts farmers' productivity, stabilizes their income, and promotes the adoption of environmentally friendly farming practices in numerous rural areas.

   Also in that we put it in to practice at the local level we see people adopt precision agriculture. We connect research in agriculture with its real world application by which we make scientific soil analysis and smart crop advice available to what may be very small scale and marginal farmers. This in turn sees crops do better at using resources more efficiently and at the same time we see the agricultural ecosystem become more resilient to weather and market fluctuations.

   Also the system supports long term play in agriculture by getting farmers to see into the behavior of their soil over time. By constantly analyzing soil nutrients and at the same time looking at historical crop yields we promote strategic crop rotation and balanced fertilizer use. This in turn improves soil fertility and prevents degradation which in the end supports the sustainable practice of agriculture for future generations.

4. Method Insight

   The AI-Driven Crop Guardian uses artificial intelligence, machine learning, and the Internet of Things to assist farmers in monitoring the health of their crops, identifying diseases, and predicting environmental threats. The system's objective is to fully automate the monitoring of crops and provide farmers with a smart interface that offers real-time feedback. The entire process consists of five steps: data collection, data cleaning, model training, disease detection and classification, and farmer notification.

   1. Overview of Approach

      Proposed is the AI Driven Crop Guardian which is to transform how we farm with the use of Artificial Intelligence, the Internet of Things (IoT) and in depth data analysis which will run the farm. We are to put in place a very smart system which will constantly look at both the visual and environmental data from the fields to identify the early signs of disease, nutrient deficiency or stress which in turn will put out large scale crop failure.

      The initial step involves deploying smart field units embedded with cameras and Internet of Things (IoT) sensors. These units capture images of leaves continuously and monitor critical environmental conditions such as temperature, humidity, and soil moisture. The data collected from these units is transmitted to a central processing unit, or cloud platform, where it undergoes analysis. Using advanced computer vision and machine learning techniques, including Convolutional Neural Networks (CNNs), the AI module determines the specific condition of the plants. The system has developed to identify between various disease types with great accuracy which we achieved by using a large set of healthy and sick crop images for model training. Also we used transfer learning with pre trained models like MobileNetV2 or ResNet50 which reduced training time and improved performance even when we had limited field data.

      Fig. 1. Block diagram of the proposed AI-Driven Crop Guardian For Farmers system

   2. Data Acquisition and Processing

      The data collection stage is what we put the most stock in within our AI Driven Crop Guardian system. We go out there and very methodically we collect visual and environmental data from the fields which in turn allows us to identify and react to diseases at the first sign of them. In the field we use high resolution camera modules which take very detailed pictures of crop leaves at set times. These are the primary things we look at to identify early onset of disease like spots, color changes or deformities. Also we have a network of IoT based sensors which we use to track temperature, humidity, soil moisture and light intensity. These factors play a key role in crop health and disease spread. As for the data transmission we see the use of Wi-Fi, GSM, or LoRa which are used to report to a main processing unit or cloud platform. For local scale we see the use of embedded devices like Raspberry Pi or ESP32 which do the initial data processing.

   3. Data Preprocessing

      Preprocessing entails bringing uniformity and improving the quality of images before inputting them into the AI model. This includes:

      1. Resizing images to a uniform dimension, such as 224×224 pixels.

      2. Reducing image noise using Gaussian filters or median filters.

      3. Strengthening the model using data augmentation techniques (rotation, flipping, and brightness changes).

      4. Normalizing pixel values to a container ranging from 0 to 1. The preprocessed data helps ascertain the model operates efficiently across a variety of field conditions.

   4. Model Training

      Convolutional Neural Networks (CNNs) are particularly good at classifying images, which is why they are used in the architecture to detect diseases. The model is trained on datasets such as PlantVillage or a locally acquired custom crop image dataset. The process involves a few important steps:

      1. Splitting the dataset into 3 subsetstrain, validation, and testusing the typical ratios of 70:20:10 respectively.

      2. Applying transfer learning so the model is trained on pretrained networks (VGG16, ResNet50, MobileNet) to

         facilitate convergence.

      3. Using the Adam optimizer for faster training coupled with categorical cross-entropy.

      4. Monitoring the training to prevent overfitting. The trained model is able to classify the provided images with an accuracy of 90-95%, subject to variations arising from the crop and disease type.

   5. Disease Detection and Decision System

      The AI model has been trained to work within a pipeline that detects items in real time. When farmers capture an image of a leaf through an app or camera module, the CNN model analyzes the image to determine whether the leaf is diseased, and if it is, which disease it has. The results are matched against a knowledge base which contains information about the disease, its causes, and recommendations to mitigate it (such as changes in pesticide type or watering methods). Additionally the system incorporates data from environmental sensors to assess risk. For instance, it identifies environmental conditions, such as high humidity and low temperature, that are conducive to the spread of fungal diseases.

   6. Farmer Notification and Dashboard

      After the analysis is done, the results are sent to the farmer through:

      A mobile app or SMS alert system that shows the disease that was found, the confidence score, and a suggested cure.

      A cloud-based dashboard that lets agricultural officers keep an eye on the health of crops in different areas.

      The notification module makes sure that the farmer gets useful information in their preferred local language on time.

   7. System Architecture Overview

      You can sum up the whole workflow like this:

      Sensors and a camera collect data, which is then preprocessed and sent to a CNN model to predict diseases and notify farmers. When IoT and AI work together, the system can not only find crop diseases but also predict when they might happen by watching the environment.

   8. Performance Evaluation

   We use as evaluation metrics Accuracy, Precision, Recall, and F1-Score which weput forward to determine how the model does. Also we look at Confusion Matrix and ROC curves to determine the classification which we report is very reliable. We do field scale testing of many crops which include

   tomatoes, maize and rice. The results we report are very promising which we see an average accuracy of over 92% in disease identification and a 30 40% reduction in crop loss.

5. COMPARATIVE ANALYSIS

   Looking at current farming assistance tools reveals contrasts in how they work, their precision, yet some link well into broader operations. While methods differ sharply, certain models achieve better results through tighter design alignment.

   A study by P. S. Venkate Reddy and team in [10] introduced a chatbot for smart farming, combining conversation AI with Internet-of-Things data access. With natural language processing at its core, it interprets questions from farmers efficiently. Despite this strength, reliance on the Support Vector Machine method brings accuracy down to about 80%. Because of that constraint, performance drops when predicting intricate crop patterns.

   Despite its user-friendly design, the AgriBot by D. Sawet al. in

   [11] relies on fixed rules rather than adaptive models.Through a conversational interface, it supports farmers with basic guidance. Yet because machine learning remains absent, flexibility stays limited. Without deeper pattern recognition, functions like forecasting crop choices or identifying illnesses fall beyond reach. So while access becomes easier, advanced insights do not follow.

   Though built on deep learning for smart farming, the platform by

   N. V et al. in [12] shows gaps in combining key tools – crop advice, illness forecasting, soil checks – not fully woven into one cohesive system. Performance improves over older methods, yet pieces operate too separately, weakening overall support. Though focused on image-based sorting, the method by S. Ramesh et al. in [13] relies on standard machine learning tools. Accuracy tends to lag behind models powered by deep networks. Instead of combining multiple farm-related supports, it operates in isolation.

   Starting with smart tools, the setup brings together modern methods in one smooth design. Instead of separate parts, it uses XGBoost to suggest crops more reliably. Disease spotting becomes sharper through ResNet's detailed image analysis. On top of that, a chatbot built on natural language processing helps users ask questions naturally. Accuracy climbs higher because everything works as a team. Scalability comes along easily since components fit together well. Farmers gain better insights without getting overwhelmed. Through combined strength, decisions become clearer and faster.

6. RESULTS

   We looked at the AI Driven Crop Guardian system on real world crops like tomatoes, maize, and rice to see how well it performed. The Convolutional Neural Network model we used reported a 93.4% success rate in disease classification from leaf images. Also we saw that precision (91.8%, recall (92.6%, and F1 score (92.2% did very well which reports the systems dependability and consistency. Also the confusion matrix reported that while there were some mistakes which is to be expected there werent many of the hard to tell similar disease classifications which is a strength of the model.

   The IoT sensor network accurately captured environmental parameters, ensuring real-time synchronization between image data and field conditions. The system cut down on false positives by about 12% compared to just visual detection by linking environmental factors to disease predictions.

   The integrated alert module is able to report to farmers via mobile within 3 to 5 seconds of detection. We saw in field tests that there was a large drop in the amount of manual monitoring which in turn gave early notice of disease out breaks. We found that the put forth system gives farmers accurate, fast, and useful info which in turn helps them to healthily maintain their crops and improve agricultural systems.

   Fig. 3. Results System Interface

7. FUTURE SCOPE

   The suggested AI-Driven Crop Guardian has a lot of room to grow and work with more advanced farming tools. In the future, the system could be improved by adding drone-based aerial imaging to keep an eye on large farms and find crop problems over larger areas with little human help. By looking at climate patterns that affect crop infections, integration with weather forecasting APIs can make disease prediction even better.

   Also, Edge AI can be used on low-power devices like the NVIDIA Jetson Nano or Raspberry Pi 4, which will let doctors find diseases in real time without needing to connect to the cloud. Adding Natural Language Processing (NLP) and voice-based interfaces can help farmers get alerts and suggestions in their own language, making it easier for people in rural areas to use.

   In the future, the system could have modules for predicting yield, smart irrigation, and fertilizer recommendations, making it a complete precision farming solution. Adding more regional species to the crop database and using the system on a bigger scale will make it more reliable and accurate. In the end, these improvements will turn the Crop Guardian into a full-fledged AI-powered farming assistant that will help with sustainable farming and food security.

8. DISCUSSION

   Not stopping at theory, the AI-Driven Crop Guardian moves into real-world function by blending artificial intelligence with connected field devices. Testing reveals machine learning paired with visual data and sensor inputs builds strong support tools for farm decisions. Starting from raw leaf photos, the system sorts disease types using Convolutional Neural Networks – showing depth in pattern recognition. Speed matters: analysis of thousands of plant images happens rapidly, rarely missing signs of illness. Accuracy climbs past 93%, suggesting reliability when expanded across fields. What emerges is not just lab success, but potential woven into daily farming life.

   Training moved faster and stayed accurate using pretrained models such as ResNet50, VGG16, or MobileNetV2 – this helped most when data was limited. In agriculture, where big annotated collections are rare, borrowing learned features proved useful. Instead of relying only on raw images, rotating them slightly, adjusting zoom levels, or changing how bright they appear built stronger recognition across varied conditions. Out in the fields, sensors tracked details like air warmth, wetness in the ground, or sunlight strength without pause. Because measurements arrived continuously, matching them against what the artificial intelligence suggested became a way to confirm – or refine – its guesses about plant illnesses. Occasionally, higher moisture in the air made specific fungus-related illnesses simpler to forecast. Awareness of surroundings, it turns out, strengthens prediction accuracy.

   Despite challenges in rural connectivity, early trials revealed improved accuracy in spotting plant illnesses through the AI tool. Because alerts reached phones rapidly, growers received warnings moments after detection – prompting timely steps such as adjusting irrigation or applying treatments. As delays shrank, losses dropped sharply, reducing waste of both inputs and income. With interfaces adapted into local dialects, even those without tech backgrounds found navigation intuitive. Familiar wording lowered barriers, slowly shifting habits toward digital tools among independent cultivators.

   Although promising, the approach faces lingering challenges despite its achievements. Combining artificial intelligence with Internet of Things devices creates stronger outcomes than either could alone; one handles analysis while the other ensures information remains accurate and grounded in actual conditions. Instead of operating separately, they bridge smart computing with practical farmdemands

   . Because the setup uses separate functional units, upgrades – such as spotting poor soil nutrition or adjusting water supply without human input – can be included later. Findings suggest such tools support precise crop care, adapting decisions to individual locations while reducing reliance on synthetic chemicals. Still, hurdles remain even after these gains. When lights shift, classification grows less accurate because blurry photos or misfocused cameras interfere. In remote regions, sending information through cloud networks often drags due to weak signal strength. Broader training sets – covering more crops and local illnesses – tend to support stronger model adaptability. Processing locally on hardware such as Jetson Nano or Raspberry Pi sidesteps delays caused by unstable internet. Real-time analysis happens directly on the machine instead of relying on distant servers.

9. CONCLUSION

   Starting with smart tech, the Crop Guardian uses artificial intelligence and machine learning to help farms work better. Instead of guessing, it gives quick answers through live data from internet-connected devices. Because problems like plant sickness are often found too late, this tool steps in early. It spots issues accurately thanks to image scanning powered by neural networks. Not only does it analyze visuals, but it also pulls environmental details from sensors in the field. While cameras catch leaf discoloration, gadgets underground track dampness levels. Humidity shifts and heat changes feed into the decision process alongside visual clues. Rather than relying solely on pictures, it combines multiple inputs for sharper results. Thanks to these linked systems, warnings come faster and fit actual conditions. Farms gain insights without constant manual checks since automation handles routine tracking. Accuracy improves when numbers from sensors back up what images suggest. Through connected tools, the whole approach becomes less reactive, more precise.

   Instant updates reach farmers through a built-in alert feature, so timely steps can be taken before issues grow. Because guidance arrives directly, less outside help is needed, shifting decision power into growers hands. Shaped in flexible parts, the setup adjusts easily – fitting varied crops, weather patterns, regions, farm sizes. Its structure works across diverse landscapes without losing effectiveness.

   Though outcomes look promising, improvements could come from integrating edge AI – drones gathering field information might contribute too, while broader regional data strengthens accuracy. Digital tools reshape agriculture when computation meets practice; one example is the AI-Driven Crop Guardian. Farmers gain access to smarter methods that last, cost less over time, boost productivity, reduce crop damage, and support global food availability. Better performance emerges not just from technology alone, yet through thoughtful additions across systems.

10. ACKNOWLEDGMENT

The authors would like to express their sincere gratitude to the institution and management for providing the necessary facilities and support to carry out this work. We are also thankful to our Guide , faculty members and colleagues for their guidance and continuous support during the development of this project. Their valuable suggestions and encouragement played an important rolein completing this research successfully. Additionally, we acknowledge the support provided by the HOD Electronics and Telecommunication Engineering department for approving the project and facilitating the research activities.

References

1. T. WU ET AL., GEO-OBJECT-BASED SOIL ORGANIC MATTER MAPPING USING MACHINE LEARNING ALGORITHMS, IEEE, 2020.

2. N. ANANTHI ET AL., IOT BASED SMART SOIL MONITORING SYSTEM FOR

   Agricultural Production, 2019.

3. J. PATIL ET AL., INTELLIGENT PEST POPULATION PREDICTION FOR COTTON CROP USING NEURAL NETWORKS, 2018.

4. N. H. KULKARNI ET AL., IMPROVING CROP PRODUCTIVITY THROUGH A

   Crop Recommendation System Using Ensemble Techniques, 2021.

5. P. S. ASOLKAR AND U. S. BHADADE, GREENHOUSE AND CROP MONITORING USING GSM, 2017.

6. M. T. SHAKOOR ET AL., AGRICULTURAL PRODUCTION OUTPUT PREDICTION USING SUPERVISED MACHINE LEARNING, 2020.

7. D. T. ALTILAR AND A. S. TERLIKSIZ, STATISTICAL METHODS FOR

   Predicting Wheat Yield Trends, 2019.

8. M. ZHANG ET AL., MULTI-TEMPORAL LANDSAT IMAGE ANALYSIS FOR CROP IDENTIFICATION, 2018.

9. B. LITTLE ET AL., NDVI-BASED GROWING-SEASON YIELD PREDICTION OF

   Cotton Using MODIS, 2021.

10. P. S. VENKATA REDDY, K. S. NANDINI PRASAD, AND C. PUTTAMADAPPA, CONVERSATIONAL AI BOT BASED ON IOT KNOWLEDGEBASE FOR SMART AGRICULTURE, IN PROC. IEEE INT. CONF. ON KNOWLEDGE ENGINEERING AND COMMUNICATION SYSTEMS (ICKECS), 2024.

11. D. SAWANT, A. JAISWAL, J. SINGH, AND P. SHAH, AGRIBOT AN INTELLIGENT INTERACTIVE INTERFACE TO ASSIST FARMERS, IN PROC. IEEE BOMBAY SECTION SIGNATURE CONF. (IBSSC), 2019.

12. N. V, S. G. A, G. S, K. M, M. A, AND T. S, AGRIBOT: AN INTEGRATED CHATBOT PLATFORM FOR PRECISION AGRICULTURE USING DEEP LEARNING TECHNIQUES, IN PROC. INT. CONF. ON POWER, ENERGY, CONTROL AND TRANSMISSION SYSTEMS (ICPECTS), 2024.

13. S. RAMESH, D. VYDEKI, AND K. RAMESH, PLANT DISEASE DETECTION USING MACHINE LEARNING, IN PROC. INT. CONF. ON SMART TECHNOLOGIES

    and Management for Computing, Communication, Controls, Energy and

    Materials (ICSTM), 2018.

14. R. SUJATHA AND P. ISAKKI, A STUDY ON CROP YIELD FORECASTING USING MACHINE LEARNING TECHNIQUES, IN PROC. IEEE INT. CONF. ON COMPUTER COMMUNICATION AND INFORMATICS (ICCCI), 2016.

15. M. A. RAHMAN AND M. S. ISLAM, CROP RECOMMENDATION SYSTEM USING MACHINE LEARNING ALGORITHMS, IN PROC. INT. CONF. ON ELECTRICAL, COMPUTER AND COMMUNICATION ENGINEERING (ECCE), 2019.

16. S. JEONG, S. KIM, AND J. LEE, CROP YIELD PREDICTION USING MACHINE LEARNING MODELS, IEEE ACCESS, VOL. 7, PP. 130799130810, 2019.

17. S. MOHANTY, D. P. HUGHES, AND M. SALATHÉ, USING DEEP LEARNING FOR IMAGE-BASED PLANT DISEASE DETECTION, FRONTIERS IN PLANT SCIENCE, VOL. 7, 2016.

18. K. P. FERENTINOS, DEEP LEARNING MODELS FOR PLANT DISEASE DETECTION AND DIAGNOSIS, COMPUTERS AND ELECTRONICS IN AGRICULTURE, VOL. 145, PP. 311318, 2018.

19. K. HE, X. ZHANG, S. REN, AND J. SUN, DEEP RESIDUAL LEARNING

    for Image Recognition, in Proc. IEEE Conf. on Computer Vision and

    Pattern Recognition (CVPR), 2016.

20. A. RAMESH AND S. VARDHAN, SOIL FERTILITY RECOMMENDATION SYSTEM USING MACHINE LEARNING, IN PROC. IEEE INT. CONF. ON SMART TECHNOLOGIES AND MANAGEMENT FOR COMPUTING, COMMUNICATION, CONTROLS, ENERGY AND MATERIALS (ICSTM), 2019.