Information leakage-aware computer aided cyber-physical manufacturing
A methodology as described herein allows cyber-domain tools such as computer aided-manufacturing (CAM) to be aware of the existing information leakage. Then, either machine process or product design parameters in the cyber-domain are changed to minimize the information leakage. This methodology aids the existing cyber-domain and physical-domain security solution by utilizing the cross-domain relationship.
1. A method of protecting a cyber-physical additive manufacturing (“AM”) system from information leakage, wherein the AM system comprises a cyber-domain, wherein a G-code resides in the cyber domain, wherein the G-code comprises a set of design variables, and a physical domain, comprising a manufacturing machine wherein the manufacturing machine has a plurality of side channels, wherein the plurality of side channels have analog emissions, wherein the analog emissions depend on control signals, wherein the G-code comprises an instruction to the manufacturing machine, wherein the G-code is converted by the cyber domain into a set of the control signals in the physical domain, wherein the control signals depend on the set of design variables of the G-code, wherein the execution of the control signals causes the AM system to emit the analog emissions in the plurality of side channels in the physical domain, wherein physical properties of the analog emissions in the side channels depend on the composition of the control signals being executed, wherein the set of design variables influence the physical properties of the analog emission, the method comprising:
(a) obtaining a benchmark G-code, wherein the benchmark G-code has an initial set of values for the set of design variables;
(b) obtaining an estimated G-code for the initial set of values for the set of design variables by:
i. executing the benchmark G-code on the AM system, wherein the benchmark G-code is converted to the control signals, wherein the control signals are executed by the AM system;
ii. measuring the analog emissions from the plurality of side channels while the benchmark G-code is executing;
iii. repeating steps (i) and (ii) for a plurality of benchmark G-codes and corresponding control signals, while training a supervised learning algorithm to learn an estimation function, wherein the estimation function estimates control signals of each G-code of the plurality of benchmark G-codes from a plurality of analog emissions:
iv. executing at least one subsequent G-code on the AM system:
v. observing at least one subsequent set of analog emissions from the plurality of side channels:
vi. using the estimation function to estimate at least one subsequent set of control signals from the at least one subsequent set of analog emissions; and
vii. computing an estimated G-code from the estimated at least one subsequent set of control signals;
(c) using the benchmark G-code and the estimated G-code to compute a mutual information value for the set of design variables, wherein the mutual information is a measure of how accurate the estimated G-code is compared to the benchmark G-code;
(d) modifying the set design variables of the G-code to obtain a modified G-code and a modified set of control signals;
(e) repeating steps (b) through (d) a plurality of times, to obtain a plurality of mutual information values for each of a plurality of sets of design variables, wherein the plurality of sets of design variables spans the set of design variables;
(f) selecting the set of design variables corresponding to the minimum mutual information, to obtain an optimal set of design variables;
(g) for each future G-code to be executed on the AM system, modifying the set of design variables to the optimal set of design variables to obtain a leakage optimized G-code; and
(h) executing the leakage optimized G-code on the AM system.