TITLE: "Drone Doctors: Deep Learning of plant topography for predictive / prescriptive analysis against pest and diseases".
BACKGROUND: Crop losses owing to pest & diseases are inherent in Indian agriculture with the annual loss of 15-25% of productivity. Pest & diseases are complex, crop/region specific, seasonal, epidemic/endemic, which require integrated approaches to manage the loss. Due to the level of complexity, diagnosis for preventive measures are challenging, particularly our inability to (fore)see the pest/disease occurrence and their life cycle, while the level of difficulty raises with the size of land holdings. Due to the poor visibility of pest and disease occurrences, our ability to integrate and use the data for preventive/prescriptive measures has been the challenge leading to continuous productivity loss.
PROBLEM DESCRIPTION: Crop loss is generally diagnosed by the symptoms developed by pest and diseases, while manual diagnosis has a very limited scope in identifying the damaged plant parts, while recognizing the pattern of pest/disease distribution almost close to impossible in large area. Drone based scanning the large area using both visible and near-infrared light, could help track changes in plants at large scale and identify the distribution pattern very precisely. Further, an artificial intelligence (AI) driven software could process the images; integrate the existing knowledge and develop the solutions in real-time. In addition, such approach could also open the scope of predictive analysis to execute automate alert to the farmers, potentially improve the crop productivity to more than 20%.
PROBLEM STATEMENT: To build a custom drone and integrated software with AI for predictive/prescriptive model against pest and diseases in crop plants.
The proposed requirements are as follows,
1. Drone integrated with IR lenses
2. Communication device for data transmission to cloud
1. “Intelligent” Software that process the digital images; extract the binaries; categorize them to diagonise the biotic stress; integrate with existing knowledge; execute the predictive / prescriptive analysis and generate alerts for communication.
2. Cloud environment with management dashboard
1. An aerial survey and data collection system, fully functional with visible and IR lenses.
2. Managed communication of the images/data to the cloud.
3. AI driven Software – fully functional for predictive and prescriptive analytics
4. Alert system for users (including farmers) in various formats including RSS feeds (for news channels) and mobile applications.
TITLE: Time Series Machine Learning based Prediction Model to achieve consistent Product Moisture in manufacturing
BACKGROUND: Achieving consistency in product quality is one of the key objectives in a manufacturing process. The product quality in a process is affected by both, process and environmental parameters. In case of a hygroscopic organic product, moisture plays a vital role in determining the end product quality. In the case of this problem it is crucial to maintain the end product moisture level within a narrow band for superior product quality. To achieve the same, online control of various process parameters is required in manufacturing.
PROBLEM DESCRIPTION: The manufacturing process of this product has a drying step which is controlled to get the end product moisture within limits. However, post drying, the retention time and conditions in further process steps impact the final product moisture. The incoming material loses moisture in the drying process. The dryer process parameters are controlled through a PID loop with short and long term levers. Post drying, the product is stored in a conditioned environment and is transferred for manufacture as per requirement. Product interacts with the surrounding environment while residing in the storage area and the manufacturing process. Due to these interactions, the product moisture changes from the desired value of 13.5+/-0.3%. Building a time series machine learning model to predict Ex- Dryer moisture set point will help to achieve Ex-Packing moisture within 13.5+/-0.3% with maximum accuracy.
PROBLEM STATEMENT: To build a Time series Neural Network (LSTM/GRU) model to predict Dryer set point in order to achieve Ex-packer moisture within 13.5+/-0.3
1. To develop a time series machine learning model to predict Dryer moisture set point in order to achieve Ex-packer moisture within 13.5+/-0.3 %
a. To counter the variation of the ambient & storage conditions, the target variable (Ex-Dryer Moisture) is changed to bring the final product moisture (var13) in the spec range of 13.5+/-0.3%.
2. Desired Accuracy Level (For Training and Model Learning):
96% accuracy has been achieved. Further improvement is desired to achieve accuracy of 98+% levels
3. Desired Accuracy Level (For Testing):
The built model should be tested on data points which have Ex-packer moisture (var13) within 13.5+/-0.3 %, predicting the target dryer set point value as close to actual set point value with an accuracy of 99%+.
TITLE: Leveraging Artificial Intelligence & Image Processing for Online Inspection of Packed Cases
1.BACKGROUND: Leaves of a crop are threshed and are packed into 200 Kg cases. 10% of these packed cases are later inspected for conformity to the master case approved by the customer in terms of Color, Ripeness and Uniformity. The quality inspection processes are manually operated and rely on the judgmental experience of the experts. The judgment is heavily driven by personal, business and environmental factors and is highly subjective.
2.PROBLEM DESCRIPTION: Inspection is a crucial activity to ensure customer satisfaction. Although it doesn’t eliminate the defects in the product, it helps identify the defective products before they are dispatched to the customer. The limitations with the existing inspection process is multi-fold.
• While Customer expects all the cases to be inspected, due to space and man-power constraints, today, the business is able to achieve only 10% inspection.
• As the inspection process happens one day after the cases are processed, due to limitations with Expert availability, real-time corrective actions in the factory in case of deviations in product quality gets difficult
• Due to human involvement in the visual inspection, there is inherent subjectivity involved in the process
3. PROBLEM STATEMENT: To automate in real-time, the packed case inspection using Machine Learning and Image Processing techniques and enhancing the objectivity of the inspection process.
4. PROJECT DELIVERABLES:
(a) To develop 3 separate algorithms which imitate Color, Ripeness and Uniformity inspection while keeping the processing time for each of the algorithms under 1 minute each
(b) The developed Algorithms to be generic of the grade type for all Color, Ripeness and Uniformity dimensions
(c) To identify patterns for Color, Ripeness & Uniformity inspection, if any by understanding the way Algorithm is functioning
TITLE: Packaging format and design optimizer for packaged food products
1. BACKGROUND: The packaging format design is currently based on trial and error for most of the packaged food products. The packaging format decides the damage any food product will undergo, especially during transit and handling. Especially for biscuits, product breakage is a prime concern, and any packing format is decided retrospectively based on the damage endured during transits (mostly by evaluating breakage). A simulation based model would solve this problem, just like any mechanical design, if packaging design can be tested for relative preference.
2. PROBLEM DESCRIPTION: Packaging format for food products need to be designed based on a simulation model. This model should involve a multi body dynamics study to evaluate the maximum forces the final packaged food will be subject to during transit and handling. A simulation has to be done basis forces that products undergo during the journey in the supply chain.
3. PROBLEM STATEMENT: To build a computer simulation model for optimizing packaging design for a given packaged food
4. PROJECT DELIVERABLES:
(a) A comprehensive stimulation model which will help in designing the optimized packaging format.
(b) The simulator should identify and recommend the areas that need to be strengthened for a given load distribution so as to minimize the effects on the product.
(c) The model has to be made modular taking inputs from different aspects of transit in the supply chain and should be able to be generalized to any product.
TITLE: Intelligent Food Packaging: Time – Temperature Sensors to deliver Super-Safe Frozen Shrimp Meat Products
1. BACKGROUND: To deliver highly relished and supersafe prawns to the consumer, rapid freezing after processing and storage at low temperatures is essential for prawn and prawn products in order to eliminate oxidation, denaturation of proteins, sublimation and recrystallization of ice crystals. Temperature increase of frozen product will result in off-flavors, rancidity, dehydration, weight loss, and loss of juiciness, drip loss, toughening, microbial spoilage and autolysis of the product. Therefore, it is absolutely critical that frozen conditions are maintained during transportation, storage, and in retail stores. This needs to be verifiably demonstrated to the consumer during the time of purchase. This is feasible only by a bio-sensor which will demonstrate the actual state of the product during its shelf-life.
2. PROBLEM DESCRIPTION: Prawns may be frozen–thawed many times before being consumed. The Frozen prawns may get thawed (core temperature ~ 0 deg C) and refrozen during transportation and repackaging. An increase in the freeze–thaw cycles, greater than -180C, results in augmenting Thiobarbituric acid (TBA) value and cutting force while lowering the salt soluble protein (SSP) value. This leads significant deterioration in product quality. The prawns can be freeze–thawed only up to 2 cycles. In addition, the product will be spoiled if there is a slow re-freezing time which steeply increases the bacterial count in the prawns. Therefore, it is important demonstrate the temperature fluctuations and actual state of the monitored packed frozen meat to the consumer illustrating the product quality and safety.
3. PROBLEM STATEMENT: To build a low-cost Time Temperature visual biosensors which will change it basic characteristics, such as color, and ensure ongoing communication with shoppers. This should result in maximizing freshness, minimize waste, and augment the brand value of frozen shrimp products.
4. PROJECT DELIVERABLES
(a) To develop a time series data logger which will be both visual and machine readable. It should be user friendly, can be calibrated to suit products with varying shelf-life, and compatible with existing other machine-logging technologies
(b) Desired protypes (For Testing): The built prototype should be accurate, precise, and reproducible in temperature monitoring, especially with increasing temperature from – 18 deg. C., to ambient condition. Changes in characteristics, such as color (or) chemical properties, should be faster with increasing temperature and preferably capture the entire shelf-life. Integration of technology using existing raw materials. process, and products, such as paper and films, and printing will be added advantage