Artificial Intelligence (AI) in the Indian Defence Sector – Concluding Part

BY KAVITA NAGPAL

Taking forward from our previous blog on ‘Artificial Intelligence (AI) in the Indian Defence Sector’ https://thedefencespace.com/2023/07/07/part-7-of-the-series-artificial-intelligence-ai-in-the-indian-defence-sector/ in this blog post we will be covering few more Artificial Intelligence (AI)-Technologies under – Process Flow Automation for Large System, Simulators / Test Equipment and Speech / Voice Analysis Systems using Natural Language Processing

All of these mentioned AI-Technologies are being indigenously developed or developing by the India’s Defence Industrial Base (DIB).

Read on to know about these in detail. Also, refer image.

I. Process Flow Automation for Large System

1.AI-Based Automation of Water Sprinkling System

The AI-based Water Sprinkler Controller will monitor dust concentration in open cast mines by IR camera, process the data using Artificial Neutral Networks and actuate the output to switch ON the corresponding solenoid driver to do the desired water Sprinkling operation (Auto mode).

In manual mode, it takes voice commands from the water sprinkler equipment operator and actuates the output to switch on the corresponding solenoid driver.

The operator can also use front panel keys to switch ON the required water sprinkling operation in Manual mode.

There is a provision for Digital I/O module to monitor safety parameters of the system and to protect the pumps.

Key features:

Voice command provision for Manual Mode operation.
Various safety logics for protection of water pump and hydraulic motor.
Operator independent system for effective operation of water sprinkling which in turn saves water, pump, and hydraulic motor.
Provision for IoT (Internet of things) for remote monitoring and configuration through Bluetooth Data Hub, WIFI/LAN (GSM/CDMA module).

The system consists of following components:

Controller with inbuilt camera
Wire harness

Unique Features:

The monitoring of dust concentration in open cast coal/ iron ore mines by using Artificial Neural Network and IR-based camera.
The system will monitor the real time dust condition based on Neural Network by camera and actuate the solenoids to perform respective operation.
Complete stack – hardware and software – designed and built in India

Provision for expansion of the relevant file name of alarm video, etc.

Provides real-time analogue output.
The system can store the driver’s real-time driving behaviour into a high-speed TF.

2. Artificial Intelligence-Based Lighting Control system on HEMM (Heavy Earth Moving Machinery)

The Smart Lighting control system is an Artificial Intelligence (AI)system to control lighting circuits based on the light present in the surrounding.

Depending on the ambient condition or poor visibility, the system senses and decides to switch the lights ON or OFF, without operator intervention.

The system uses AI for human detection and turns on the equipment lights when a person approaches the equipment to ensure adequate illumination for operator to climb up to the equipment safely.

The AI system running on the 6 cameras engineered around the machine can perform the following actions:

Depending on the lighting conditions control the lamps to ensure proper illumination with minimum power consumption.
Detect the light and visibility condition in the surrounding environment and automatically control the lamps to ensure safe operations for the operator.
Detect the presence and distance of a person behind the machine especially in the blind spot areas to ensure the safety of people working in proximity of the machine.
AI is also used in the seamless stitching of the videos coming from 6 cameras to create a 360-degree view.

Key Features:

Automatic control of lights based on weather conditions and human presence.
Detect humans and their distance from the machine with a high level of accuracy and autonomously control the machine.
360-degree bird-eye view around the machine in a single frame.
The operator is informed about the Violations/Error reported by the system or the machine via a vernacular audio interface.
The system also provides continuous monitoring of machine bucket and boom operations through real time streaming of video feeds/images with date and time stamp to check / reduce the abuse of bucket attachment during operation.

3. AI-Enabled Weld Inspection Machine with Computerized Radiography-(AI-RT)

An AI-enabled RT tool is the replacement of X-ray film technique that is used currently in conventional radiography for weld inspection.

This is a safe, rapid, reliable simulation-assisted automation which is developed using Artificial Neural Networks (ANN)/Support Vector Machines (SVM)/Convolutional Neural Network (CNN) for automated defect recognition on the Digital X-ray images.

Unique Features:

No darkroom conditions, chemical or consumables is required
Short processing time
Digital films can be stored in computers, cloud or remote network servers
Efficient archiving of the data with no degradation of the quality of the image.
Digital films cannot be tampered, ensuring more reliable data
Very cheap considering the re-usability
Environment friendly

Salient Features:

The product is usable in any steel fabrication Industry making it suitable for the domestic and international market.
Considering the time saving in processing the information with AI enablement, the product is highly recommended for usage in shipbuilding and any other steel fabrication Industry.

3. Artificial Intelligence-Enabled Weld Inspection Machine with Advanced Phased Array Ultrasound Technique-(AI-UT)

This AI-enabled UT tool employs a safe and rapid advanced phased array ultrasound technique using Full Matrix Capture/Total Focusing Method approach (FMC/TFM) tools. It is augmented with AI data analytics.

This can replace X-ray techniques that are used generally for weld inspection.

Phased Array Ultrasonic Testing (PAUT) is employed in this tool, which is an advanced nondestructive examination technique that utilises a set of ultrasonic testing (UT) probes made up of numerous small elements, each of which is pulsed individually with computer-calculated timing.

Main Functionalities:

Portable
Able to detect and obtain the images of defects based on the physics of ultrasonic waves in welded structures
Perform robotic weld inspection in area where accessibility is limited. This reduces the requirement of scaffoldings and saves time & cost
Perform automated evaluation of defect detection using AI-based tools
Safe and quick
Ease for test set-up
Rapid inspections with digital recording of data

Salient Features:

The product is usable in any steel fabrication industry making it suitable for the domestic and international market.
Considering the time saving in processing the information with AI enablement, the product is highly recommended for usage in shipbuilding and any other steel fabrication Industry.

4. AI-Based Automated Bore Cleaning

The performance and life of artillery is both important and critical and as such, weapons have to be ready 24×7.

This AI based technology facilitates an automated inspection and cleaning of large calibre artillery and tank guns.

An AI-vision enabled, robotic cleaning system is available for large guns.

Unique Features:

Fully automated robotic cleaning system.
AI-assisted vision inspection of bore surface
Designed to operate in harsh environments
Flexible cleaning system for riffled and smooth barrels
Adaptable as per gun service stage
Effortless and effective cleaning
Can run on any power outlet or battery, ensuring flexibility.
AI vision can detect and localize surface defects and unclean patches.

5. Brainbox

The structural integrity of gun barrels, missile launchers and rocket launcher barrels have a direct impact on the safety of men and equipment of the Armed Forces.

At present, akin to other material structures there is no method which quantitatively measures the fatigue of such equipment and predict residual life.

Brainbox Spica is AI technology that can be customised to undertake structural integrity analysis of various types of metal structures and fittings.

For e.g., it can assess the structural integrity of various critical components of Guns of multiple calibres and also rocket Launcher Barrels and their components as these are subject to high levels of stress and temperature variations.

The current form of analysis can only measure thickness at random locations, pitting, etc.

It cannot detect subsurface cracks or predict residual life based on actual wear and tear.

Further, it entails change of barrels on firing of a specified number of rounds, irrespective of their condition whether good or bad.

Also, it means continued usage of barrels that may have experienced excessive damage and may be unsuitable for use.

Brainbox Spica can be utilized to assess the structural integrity and anomalies of barrels and components of weapons systems at the time of installation and during routine inspection. Useful for all the armed forces.

Benefits:

High Battle readiness
Spares can be forecasted and budgeted accordingly based on criticality of failure
Only Good usable Condition Ordinances & Munitions will reach the field ensuring complete preparedness in warfare.
Reduce the overall cost of maintenance of ordinance & munitions (even other assets).
The Prescriptive data will help improve the specifications of initial asset acquisition and/or defense production based on the real useful life information of the assets.
Defense Exports, as SPICA can be exported to all friendly Nations & allies

Unique features:

It’s a deep learning system with mounted AI Algorithms for detection of metallic cracks and anomalies
Real-time inspection and assessment capability
Detects anomalies of structures such as cracks, dents, erosion, etc.
Detects anomalies of structures such as surface cracks, dents, erosion, etc.
Can detect and automatically reject components which are unacceptable or unhealthy
Reduce Downtime and Increase availability of the asset with periodic health assessment
Eliminates intuitive inspections
Can be deployed at production or maintenance stage based on the place of usage

II. Simulators / Test Equipment

1.Development of Artificial Intelligence-based training modules for technicians for operation and maintenance of SU – 30 MKl aircraft

Parallax Labs is currently working with HAL to build a Virtual Reality & Artificial Intelligence powered training solution for operation and maintenance of Sukhoi SU-30 MKI Aircraft.

The solution shall enable the operators to take up the most challenging procedures for hands-on virtual training with real-time rectification and performance analytics.

Besides, the AI model accurately evaluates the corrective repair actions taken by the trainee while comparing it with historical data to deliver compelling performance reports.

III. Speech / Voice Analysis Systems using Natural Language Processing

1.AI-Based Mandarin Translators

In order to develop indigenous capability, the Indian Army has been focusing on multiple domains to leverage AI.

In the domain of Natural Language Processing (NLP), it has focused on an AI-based Bi-directional Mandarin to English Wearable Translation device.

Device Capability

The device is a wearable device for speech-to-speech translation.

It equips intelligence agencies with a bi-directional speech-to-speech system that accepts input audio/speech in one source language (e.g., English/Mandarin) and converts it into an equivalent audio/speech of the target language (e.g., Mandarin/English).

The portable translation device has real time bi-directional speech capability, without any rearward network connectivity.

It weighs 80-100 gms, with a latency of 3-4 secs and a battery life of 12 hrs.

The system consists of wearable devices, docking stations and re-training servers.

Furthermore, it is the wearable translation device on the tactical edge and completely disconnected from any network.

The potential usage is during meetings, engagements etc.

2. AI-Based Offline Language Translator

It is the only solution in India wherein a GPU-based algorithm can be deployed in a CPU-based hardware. Further, being offline, it does not entail any kind of dependency.

Its ability to be deployed in a CPU-based system algorithm, means it can be used for field trails, or in a remote location through a hand-held mobile device.

3. Speech-to-Speech Translation

The Speech-to-Speech translation is a difficult problem to solve due to understanding of how different people speak with different clarity, accents, speed etc. compounded by noise from ambient environment.

This problem is made more challenging by requirement of huge corpus required to cover large vocabularies for translation to give a good quality output.

The Speech-to-Speech translation system developed for English to Hindi translation is able to perform well under such conditions.

The objective of the development of Speech-to-Speech translation system is to generate real- time speech translation in the target language. The system will enable the understanding of source language without the requirement of a translator.

This will reduce the requirement of translators and reduce cognitive load put on translator. Since the system works in real time it will capable of enabling communication also.

The objective of the English to Hindi Speech to Speech translation system is to enable Hindi speaking people (528 million) who are unable to understand English.

Notably, the system can be deployed in places where interaction between people of different language speaking regions is frequent. Also, it can be used to understand recorded speech in different language.

Since, the system is developed in a modular fashion, the modules of speech recognition, translation and generation of speech can also be used independently. Thus, serving different use cases.

Unique Features:

No internet connectivity required
No dependence on third-party APIs or cloud services
Comprehension of Indian accents
Generation of natural Hindi speech

4. AI-Enabled Voice Transcription Software

Being able to intercept and identify audio data, particularly voice, can be vital to security and surveillance.

This software is an integrated module of the Communication Intelligence (COMINT) system.

The Speaker and Language Identification module serves as a definite value addition towards voice surveillance on audio data intercepted by the following:

The solution is deployed at the pilot site (Jamnagar) of Signal Intelligence (SI) Navy.

It works uniquely by applying speaker and language recognition functionalities to explore the capability of revealing significant insights into the dynamics of hidden communication.

Particularly, the speaker and language details identified by the system can easily be targeted by law enforcement and security agencies for analysing covert networks.

Innovative features:

Voice activity detection
Speaker identification
Language identification
Keyword spotting
Continuous speech conversion
Speech Enhancement/ Audio Editing
Dialect Identification
Speaker Diarisation

5. Voice Activated Command System (VACS)

Modern Combat aircraft consists of various sensors like RADAR, Data Link, EO Pod etc., which provide a large amount of information in real time.

Pilots have to fly their missions and simultaneously correlate all the sensor information in real time for decision making.

VACS recognises the voice commands of the pilot and send the recognised voice command codes to the mission computer for carrying out the Pilot’s intended action.

VACS will be used for tuning of radios, selection of modes, setup of MFDs, navigational equipment etc.

Unique Features:

Voice Activated Command System recognises pilot voice commands using AI-based Speech Recognition technology. It is the first of its kind to be used in a cockpit of an airborne platform.
It can correctly recognise voice commands even with the cockpit noise of 110 dB (loudness equivalent to a riveting machine or a jackhammer) Utility
Only a limited numbers of controls are available on the Hands-On Throttle and Stick (HOTAS) to enable the pilot to fly the aircraft and simultaneously command the sensors or initiate actions without removing his hands from the flying controls.
AI-based Voice activated command system (VACS) will help to reduce pilot workload and enhance effectiveness in combat situation and allows pilot to focus more on the critical tasks such as engaging threats, landing, terrain etc. as movement of hand from HOTAS to Display devices will be eliminated.

Marketability & Export Potential

VACS has been validated and integrated on Hawk-i platform.
It is proposed to be integrated in Su-30 Upgrade and AMCA aircraft as part of Integrated Audio Management System (IAMS)
VACS can be packaged with the fighter platform sales being targeted by HAL to friendly countries.

6. AI-Powered Language Translation Platform

The BLADe-S ISR platform is an indigenously built, fully-integrated Artificial Intelligence (AI) based platform capable of processing all major forms of intelligence data – audio, video, text and images.

The platform enables intelligence analysts to review input data, analyse and then tag intercepts based on criticality. An in-built reporting engine provides custom reports to senior leadership enabling data driven decisions.

BLADe-S is the wearable verison of the system – an AI based offline real time wearable language translation platform powered by Edge computing devices with AI speech analytics and NLP.

Built in Neural netwrok processor with fedarated learning and on field adaptation capabilities Due to small footprint, zero connectivity the device is almost unobtrusive.

The device can be used by intelligent agencies and frontline positioned units to decode the foreign language messaging.

Unique features:

Wearable language translator with AI speech analytics and NLP
Built in Neural Network Processor to compute & offer real time translation using Edge computingdevice
Federated Learning (Distributed & Private – no movement of Data) & Adaptation on Field
2 mode operation- Unidirectional (Mandarin to English) & Bidirectional (Mandarin to English and viceversa)
Supports Russian and other 11 local languages
Lithium-ion battery 6-8 hours back-up in standard & harsh operating conditions
Memory: Can record up to 400 hours of data. 10 seconds of voice gets converted into a voice file of approx. size few kbs
For full day recording- Approx. data size is 500-600mb
Total memory space available is 64 Gb SC Card- 43 Gb is for application. 19 Gb available for recorder
Memory full- 8/12 hrs memory dump can be done2