India’s growing research potential in control theory and implications for flight control
India has produced many giants in the field of control theory who have contributed immensely in various branches of this wonderful discipline. Three names that immediately come to my mind are:
(1) Kumpati Narendra (Citations ~ 44000)
Kumpati Narendra - Google Scholar
(2) Shankar Sastry (Citations ~104899)
Shankar Sastry's Home Page (berkeley.edu)
Shankar Sastry - Google Scholar
(3) M Vidyasagar (Citations ~37000)
Mathukumalli Vidyasagar - Google Scholar
Out of these, Sastry has the honour and privilege of contributing the most in the field of adaptive control. He has authored several highly citated papers on stability proofs of various adaptive control formulations. Incidentally he is also an alumni of IIT Bombay, BT ’77 batch. Adaptive control has a lot potential benefit over several other control strategies as it renders the possibility of online learning/adaptation in the face of changing plant behaviour.
In recent times, reinforcement learning (RL) has gained enough traction within the controls community. In control theory, it is often implemented via what is known as “adaptive dynamic programming”. The fundamental benefit of this strategy is that approximately optimal policies could be learnt online under various actuator constraints while adhering to Lyapunov-like stability requirements. Several Indian researchers in recent times have made their contribution to this new field such as,
1) Rushikesh Kamalapurkar - Google Scholar
2) Shubhendu Bhasin - Google Scholar
3) Jagannathan Sarangapani or S. Jagannathan or sarangapani Jagannathan - Google Scholar
4) Nitin Sharma - Google Scholar
5) Balaraman Ravindran - Google Scholar
6) Amardeep Mishra - Google Scholar
While several advancements have been made in reinforcement learning (from control theoretic perspective), the aerospace community in general hasn’t embraced the new control strategy. And this has got to do with MIL spec requirements of gain and phase margins. While RL in recent times has been made more rigorous via incorporating both stability and safety (control barrier Lyapunov functions), it still requires certain fundamental assumptions notably those associated with persistency of excitation condition. In addition, a pure adaptive controller requires large gain (learning rate) to adapt itself to lets say large error buildup, large abrupt variation in dynamics etc. It however can potentially destabilize the system by inducing unwanted oscillations especially in systems with significant noise and disturbances.
In Indian context, especially from aerospace industry, considerable effort has been paid to Gain Scheduling-based PID and Model Reference Adaptive Control (MRAC). The prime flight control system of LCA Tejas consists of these two techniques.
Designing And Testing Flight Control Laws Of Light Combat Aircraft [LCA] Tejas [Aero India 2013] - YouTube
The video describes some of the challenges CLAW faced while designing the flight control law for LCA tejas.
03 Challenge in flight control systems Dr Vijay V Patel , ADA Bangalore - YouTube
It could be noted that Dr. Deodhare (an alumni of IIT Bombay BT ’84 batch) has pioneered the flight control development at ADA. In the video below he explains how LCA is not just an aircraft for India, but it represents maturity in various facets of aerospace engineering notably aerodynamic design (wind tunnel testing under various regimes), flight control laws, composites and avionics.
Dr Girish S Deodhare : Distinguished Scientist - YouTube
He beautifully explains as to how ADA and CLAW have created a wealth of knowledge resource within the country to undertake complex aerospace developments that are coming up.
For instance, according to him, the ADA/CLAW team could undertake controls development task for various classes of aircrafts namely:
(1) Tejas Mk2 with canards.
(2) Tailess aircraft design.
(3) Stealth AMCA with twin engine configuration.
While interacting with one of the senior controls engineer (A Ph.D. in flight dynamics and control from IISc Bangalore with more than a decade experience in flight controls), I was told that the Tejas comprises of many control loops for various purposes, some of which are, the primary 3-axis FBW, Mach hold autopilot, stability augmentation system (SAG), various other controllers to improve the damping characteristics etc. These control loops leverage some of the well-known control techniques from literature, for instance the primary 3-axis FBW uses gain scheduling. It entails finding best possible gains for linear state feedback controllers over all the possible operating points in the flight regime.
Similarly, there are certain other control loops that make use of model reference adaptive control (MRAC). In fact, there is a flight recovery controller that recovers the flight from edge case scenarios and bring it back to level flight. This one use sliding mode!
Development of proper flight controls is directly contingent upon extensive wind tunnel tests of various scale models under various mach regimes. These tests reveal the wealth of aerodynamic coefficients ranging from stability derivatives, damping derivatives to controls derivatives. A good mathematic model of an aircraft captures the variation of these coefficients with respect to angle of attack, mach no., side slip, various rates etc. Further, a good model replicates the true physics of flight in simulations thus paving way for controls design. It must be noted that wind tunnels alone are not sufficient, it must have requisite instrumentation setup to capture wide variety of aerodynamic data. India has a multitude of such wind tunnel tests ranging from subsonic, transonic, supersonic to now hypersonic with adequate instrumentation to capture these aerodynamic coefficients.
Wind tunnel test of tejas:
LCA tejas mk2 in wind tunnel tests
Wind tunnel tests of TEDBF:
Wind tunnel tests of AMCA:
In addition, India has developed a full fledged hardware in loop (HIL) system known as "iron bird test rig" that mimics the entire control system on the ground. Here is a brief video on the test rig:
One important challenge he spoke of was time delays. A well-behaved system could easily burst into oscillations and eventually instability if the time delays are not compensated well enough. For linear systems, the time delay compensation is well documented in literature and has been leveraged in the FBW design. However, for more fancy RL-based control strategies large time delays present a grave challenge.
@JamD , @Quwa , @SQ8