Patent Application
20200106785
Industrial control systems that operate physical systems (e.g., associated with power turbines, jet engines, locomotives, autonomous vehicles, etc.) are increasingly connected to the Internet. As a result, these control systems have been increasingly vulnerable to threats, such as cyber-attacks (e.g., associated with a computer virus, malicious software, etc.) that could disrupt electric power generation and distribution, damage engines, inflict vehicle malfunctions, etc. Current methods primarily consider attack detection in Information Technology (“IT,” such as, computers that store, retrieve, transmit, manipulate data) and Operation Technology (“OT,” such as direct monitoring devices and communication bus interfaces). Cyber-attacks can still penetrate through these protection layers and reach the physical “domain” as seen in 2010 with the Stuxnet attack. Such attacks can diminish the performance of a control system and may cause total shut down or even catastrophic damage to a plant. Currently, no methods are available to automatically detect, during a cyber-incident, attacks at the domain layer where sensors, controllers, and actuators are located. In some cases, multiple attacks may occur simultaneously (e.g., more than one actuator, sensor, or parameter inside control system devices might be altered maliciously by an unauthorized party at the same time). Note that some subtle consequences of cyber-attacks, such as stealthy attacks occurring at the domain layer, might not be readily detectable (e.g., when only one monitoring node, such as a sensor node, is used in a detection algorithm). It may also be important to determine when a monitoring node is experiencing a fault (as opposed to a malicious attack) and, in some cases, exactly what type of fault is occurring. Existing approaches to protect an industrial control system, such as failure and diagnostics technologies, may not adequately address these problems—especially when multiple, simultaneous attacks and/faults occur since such multiple faults/failure diagnostic technologies are not designed for detecting stealthy attacks in an automatic manner.
In addition, many current attack detection technologies are passive and rely solely on monitoring data collected from the cyber-physical system. These types of protection schemes may fail in the face of a mass spoofing attack and/or a replay attack. It would therefore be desirable to protect a cyber-physical system from cyber-attacks and other abnormal operation in an automatic and accurate manner even when attacks percolate through the IT and OT layers and directly harm control systems.
According to some embodiments, a cyber-physical system may have a plurality of system nodes including a plurality of monitoring nodes each generating a series of current monitoring node values over time that represent current operation of the cyber-physical system. A watermarking computer platform may randomly inject a watermarking signal into an injection subset of the system nodes. The watermarking computer platform may then receive current monitoring node values over time and generate a current watermarking feature vector based on the current monitoring node values. The watermarking computer platform might comprise a dedicated watermarking abnormality detection platform or a unified abnormality detection platform (e.g., that also uses data-drive feature vectors). The injection subset may be associated with a randomly selected subset of the system nodes and/or magnitudes of watermarking signals that are randomly selected.
Some embodiments comprise: means for randomly injecting, by a computer processor of a watermarking computer platform, a watermarking signal into an injection subset of the system nodes; means for receiving current monitoring node values over time; and means for generating a current watermarking feature vector based on the current monitoring node values.
Some technical advantages of some embodiments disclosed herein are improved systems and methods to protect one or more cyber-physical systems from abnormalities, such as cyber-attacks, in an automatic and accurate manner.
In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of embodiments. However, it will be understood by those of ordinary skill in the art that the embodiments may be practiced without these specific details. In other instances, well-known methods, procedures, components and circuits have not been described in detail so as not to obscure the embodiments.
One or more specific embodiments of the present invention will be described below. In an effort to provide a concise description of these embodiments, all features of an actual implementation may not be described in the specification. It should be appreciated that in the development of any such actual implementation, as in any engineering or design project, numerous implementation-specific decisions must be made to achieve the developers' specific goals, such as compliance with system-related and business-related constraints, which may vary from one implementation to another. Moreover, it should be appreciated that such a development effort might be complex and time consuming, but would nevertheless be a routine undertaking of design, fabrication, and manufacture for those of ordinary skill having the benefit of this disclosure.
As described with respect to
The system 100 may also include a watermarking computer platform 150 that automatically outputs randomly selected injection points and values and receives information from the monitoring nodes 110 to generate watermarking features. As used herein, the term “automatically” may refer to a process that may be performed with little (or no) human intervention. An abnormal boundary creation platform may receive the data-driven features and watermarking features and output one or more global boundaries (e.g., associated with data-based and/or watermarking-based feature vectors) and local boundaries for various monitoring nodes (e.g., each being associated with data-based and watermarking-based feature vectors). These boundaries can then be used to implement abnormality detection for the 100 cyber-physical system.
At S210, a computer processor of a watermarking computer platform may actively and randomly inject a watermarking signal into an injection subset of the system nodes. Note that the injection subset might be associated with a control layer physical location in a control network architecture. According to some embodiments, the injection subset is a randomly selected subset of the system nodes. Similarly, a magnitude of a plurality of watermarking signals may be a randomly selected. Note that the watermarking computer platform might utilize a Pseudo-Noise (“PN”) sequence, a Pseudo-Random Binary Sequence (“PRBS”), or any other randomization approach such that an attacker cannot predict the inserted watermark.
At S220, the system may receive current monitoring node values over time (e.g., from monitoring nodes associated with sensors and actuators located throughout the cyber-physical system). At S230, the system may generate a current watermarking feature vector in accordance with any of the embodiments described herein based on the current monitoring node values. According to some embodiments, the watermarking computer platform may include a dedicated watermarking abnormality detection platform that compares the current watermarking feature vector with a watermarking feature decision boundary and generates a watermarking abnormal alert signal. In this case, a decision fusion platform may receive the watermarking abnormal alert signal and a data-driven abnormal alert signal and generate a fused system status alert signal. According to other embodiments, the watermarking computer platform may be associated with a unified abnormality detection platform. In this case, a feature augmentation platform may augment data-driven feature vectors and at least one watermarking feature vector
In this way, embodiments may provide a method and system for domain-level dynamic physical watermarking in control systems to help ensure the integrity of the system and detect any tampering by adversaries. The watermarking mechanism may inject random perturbations through the system whose expected outcomes are known to a defender (e.g., the owner of an industrial asset) via knowledge of the underlying dynamics of the plant and the control system architecture. Both the generation scheme of the watermarking signals and the locations of the injection might be randomized to diminish any chance of being reproduced by an attacker. Moreover, the physical watermarking technology for cyber-attack detection may be provided in control systems (such as combine cycle power plants, wind power generation both at the wind turbine unit and farm level, etc.). Note that digital watermarking has been used for years in communication networks as the process of hiding digital information in a carrier signal; the hidden information should, but does not need to, contain a relation to the carrier signal. Digital watermarks may be used to verify the authenticity or integrity of the carrier signal or to show the identity of its owners. Physical watermarking, on the other hand, is an emerging technology area that may help secure cyber-physical networked control systems. In addition to a control command computed through a control law, a controller may command actuators to inject into the system a component that is random (and not known in advance) in order to secure the system against such attacks. This may be specifically useful to help detect man-in-the-middle and replay attacks such as Stuxnet. By employing physical watermarking, the covariance of the innovations signal (the difference between the measured signature of the injected watermarking signal verses the expected signature) when the system is “normal” as compared to when it is under attack are significantly different.
Because the system is excited by injecting an additional and random control signal to guard against adversarial attacks, watermarking is an “active” defense strategy. As a result, the process will impact controls cost. Game-theoretic approaches might be used to set an optimized policy to switch between cost-centric and security-centric controllers (e.g., by formulating the problem as a stochastic game between a defender and an adversary).
Also note that watermarking should not destabilize a control loop. With respect to linearized plant and controller models, system perturbations may be kept within the bP,C of the system, where bP,C is a ball of tolerable uncertainty around the nominal plant that is still stabilized by the controller. It may be characterized as the inverse of the H∞ norm of a generalized (in the sense of the robust control framework) plant. This may be useful in the sense that either the plant or the controller can be considered as a perturbed version of their nominal ones.
With respect to non-linear plant and/or controller models, the watermarking signals may be designed to keep the system trajectories within a region of attraction. If the models are available in analytical closed forms, this may be achieved through Lyapunov-based methods (otherwise, the process may be performed using numerical simulations).
If the plant and/or controller has one or more hard constraints (such as maximum magnitudes and rates of the control commands), these may also be considered. An element of the perturbation approach to authentication of assets against cyber-attack may be the development of probing signals that satisfy orthogonality conditions to permit a cross-correlation analysis to detect changes in the responses of actuators and/or sensors. To design effective random watermarking signals, PN sequences or PRBS generation schemes may be utilized. The generation of these sequences may be done in a similar fashion as in the system identification solutions (such as in closed-loop subspace identification methods). The watermarking signal design may be formulated as an inverse system identification problem. In system identification, the input signal is designed to satisfy persistence excitation property for identification of the unknown system model. Here, the system model is known, and the input signals are designed to exhibit detectable signatures in the input-output cross-correlations while having minimal effect on the normal control performance.
According to some embodiments, watermarking features may be associated with machine learning. Note that the covariance between an injected watermarking signal and the sensor outputs might be used in a statistical hypothesis test, such as an X2 detector. This method may be susceptible to high false positives and false negatives due to the challenge of the proper selection of the detection threshold. Related to this issue, a X2 detector works under the fundamental assumption that the signals follow a Gaussian distribution. Although this assumption might be valid for sensor noise, it might not be accurate for the process noise. It may also restrict the design of the watermarking signal. Some embodiments described herein may use the correlation signal as a feature for a machine learning process within an attack detector (which is trained through supervised learning or semi-supervised learning using training datasets). This technique may eliminate both the need to set a threshold on the covariance and the assumption that the signals are Gaussian (since it is a non-parametric method). Additional properties of the innovation sequence (the difference between the measured outcomes vs. expected outcomes), such as whiteness or mean value, can also be used as additional features for machine learning-based classification.
As description of how watermarking features may be incorporated in a data-driven abnormality detection system will now be provided.
Information from the normal space data source 320 and the abnormal space data source 330 may be provided to an abnormal detection model creation computer 360 that uses this data to create a decision boundary (that is, a boundary that separates normal behavior from threatened behavior). The decision boundary may then be used by an abnormal detection computer 350 executing an abnormal detection model 355. The abnormal detection model 355 may, for example, monitor streams of data from the monitoring nodes 310 comprising data from sensor nodes, actuator nodes, and/or any other critical monitoring nodes (e.g., sensor nodes MN1 through MNN) and automatically output global and local abnormal alert signal to one or more remote monitoring devices 370 when appropriate (e.g., for display to an operator or to have the global and local information fused in accordance with any of the embodiments described herein). According to some embodiments, information about detected threats may be transmitted back to a cyber-physical system control system. Note that, according to some embodiments, the abnormal detection computer 350 may implement a hybrid approach that also incorporates watermarking (and watermarking features) when making an abnormality classification (e.g., as described with respect to
As used herein, devices, including those associated with the sytstem 300 and any other device described herein, may exchange information via any communication network which may be one or more of a Local Area Network (“LAN”), a Metropolitan Area Network (“MAN”), a Wide Area Network (“WAN”), a proprietary network, a Public Switched Telephone Network (“PSTN”), a Wireless Application Protocol (“WAP”) network, a Bluetooth network, a wireless LAN network, and/or an Internet Protocol (“IP”) network such as the Internet, an intranet, or an extranet. Note that any devices described herein may communicate via one or more such communication networks.
The abnormal detection model creation computer 360 may store information into and/or retrieve information from various data stores, such as the normal space data source 320 and/or the abnormal space data source 330. The various data sources may be locally stored or reside remote from the abnormal detection model creation computer 360. Although a single abnormal detection model creation computer 360 is shown in
A user may access the sytstem 300 via one of the monitoring devices 370 (e.g., a Personal Computer (“PC”), tablet, or smartphone) to view information about and/or manage threat information in accordance with any of the embodiments described herein. In some cases, an interactive graphical display interface may let a user define and/or adjust certain parameters (e.g., abnormal detection trigger levels) and/or provide or receive automatically generated recommendations or results from the abnormal detection model creation computer 360 and/or abnormal detection computer 350.
Thus. some embodiments described herein may use time series data from one or more monitoring nodes 310 from a physical (i.e., industrial or enterprise) asset and provide a reliable abnormality detection with low false positive rate. The system may extract features from the time series data for each monitoring node. The term “feature” may refer to, for example, mathematical characterizations of data. Examples of features as applied to data might include the maximum and minimum, mean, standard deviation, variance, settling time, Fast Fourier Transform (“FFT”) spectral components, linear and non-linear principal components, independent components, sparse coding, deep learning, etc. The type and number of features for each monitoring node might be optimized using domain-knowledge and/or a feature discovery process. The features may be, for example, calculated over a sliding window with consecutive samples of specified duration from time series data. The length of the window and the duration of overlap for each batch may be determined from domain knowledge and an inspection of the data or using batch processing. Note that features may be computed at the local level (associated with each monitoring node) and the global level (associated with all the monitoring nodes, i.e., the whole asset). The time-domain values of the nodes or their extracted features may be, according to some embodiments, normalized for better numerical conditioning.
The decision boundary can then be used to detect abnormal operation (e.g., as might occur during cyber-attacks). For example,
Some embodiments described herein may take advantage of the physics of a control system by learning a priori from tuned high-fidelity equipment models and/or actual “on the job” data to detect single or multiple simultaneous adversarial threats to the system. Moreover, according to some embodiments, all monitoring node data may be converted to features using advanced feature-based methods, and the real-time operation of the control system may be monitored in substantially real-time. Abnormalities may be detected by classifying the monitored data as being “normal” or disrupted (or degraded). This decision boundary may be constructed using dynamic models and may help enable early detection of vulnerabilities (and potentially avert catastrophic failures) allowing an operator to restore the control system to normal operation in a timely fashion.
Note that an appropriate set of multi-dimensional feature vectors, which may be extracted automatically (e.g., via an algorithm) and/or be manually input, might comprise a good predictor of measured data in a low dimensional vector space. According to some embodiments, appropriate decision boundaries may be constructed in a multi-dimensional space using a data set which is obtained via scientific principles associated with DoE techniques. Moreover, multiple algorithmic methods (e.g., support vector machines or machine learning techniques) may be used to generate decision boundaries. Since boundaries may be driven by measured data (or data generated from high-fidelity models), defined boundary margins may help to create an abnormal zone in a multi-dimensional feature space. Moreover, the margins may be dynamic in nature and adapted based on a transient or steady state model of the equipment and/or be obtained while operating the system as in self-learning systems from incoming data stream. According to some embodiments, a training method may be used for supervised learning to teach decision boundaries. This type of supervised learning may take into account on operator's knowledge about system operation (e.g., the differences between normal and abnormal operation).
Since attacks might be multi-prong (e.g., multiple attacks might happen at once), DoE experiments may be designed to capture the attack space (e.g., using full factorial, Taguchi screening, central composite, and/or Box-Behnken). When models are not available, these DoE methods can also be used to collect data from real-world asset control system. Experiments may run, for example, using different combinations of simultaneous attacks. Similar experiments may be run to create a data set for the normal operating space. According to some embodiments, the system may detect “degraded” or faulty operation as opposed to a threat or attack. Such decisions may require the use of a data set for a degraded and/or faulty operating space.
Note that many different types of features may be utilized in accordance with any of the embodiments described herein, including principal components (weights constructed with natural basis sets) and statistical features (e.g., mean, variance, skewness, kurtosis, maximum, minimum values of time series signals, location of maximum and minimum values, independent components, etc.). Other examples include deep learning features (e.g., generated by mining experimental and/or historical data sets) and frequency domain features (e.g., associated with coefficients of Fourier or wavelet transforms). Embodiments may also be associated with time series analysis features, such as cross-correlations, auto-correlations, orders of the autoregressive, moving average model, parameters of the model, derivatives and integrals of signals, rise time, settling time, neural networks, etc. Still other examples include logical features (with semantic abstractions such as “yes” and “no”), geographic/position locations, and interaction features (mathematical combinations of signals from multiple monitoring nodes and specific locations). Embodiments may incorporate any number of features, with more features allowing the approach to become more accurate as the system learns more about the physical process and threat. According to some embodiments, dissimilar values from monitoring nodes may be normalized to unit-less space, which may allow for a simple way to compare outputs and strength of outputs.
where S stands for a monitoring node quantity at “k” instant of time. Moreover, output may then be expressed as a weighted linear combination of basis functions as follows:
where S0 is the average monitoring node output with all threats, wj is the jth weight, and Ψj is the jth basis vector. According to some embodiments, natural basis vectors are obtained using a covariance of the monitoring nodes' data matrix. Once the basis vectors are known, weight may be found using the following equation (assuming that the basis sets are orthogonal):
wj=(S−S0)TΨj
Note that weights may be an example of features used in a feature vector.
Thus, embodiments may enable the passive detection of indications of multi-class abnormal operations using real-time signals from monitoring nodes. Moreover, the detection framework may allow for the development of tools that facilitate proliferation of the invention to various systems (e.g., turbines) in multiple geolocations. According to some embodiments, distributed detection systems enabled by this technology (across multiple types of equipment and systems) will allow for the collection of coordinated data to help detect multi-prong attacks. Note that the feature-based approaches described herein may allow for extended feature vectors and/or incorporate new features into existing vectors as new learnings and alternate sources of data become available. As a result, embodiments may detect a relatively wide range of cyber-threats (e.g., stealth, replay, covert, injection attacks, etc.) as the systems learn more about their characteristics. Embodiments may also reduce false positive rates as systems incorporate useful key new features and remove ones that are redundant or less important. Note that the detection systems described herein may provide early warning to cyber-physical system operators so that an attack may be thwarted (or the effects of the attack may be blunted), reducing damage to equipment.
According to some embodiments, a system may further localize an origin of a threat to a particular monitoring node. For example, the localizing may be performed in accordance with a time at which a decision boundary associated with one monitoring node was crossed as compared to a time at which a decision boundary associated with another monitoring node was crossed. According to some embodiments, an indication of the particular monitoring node might be included in a threat alert signal.
Some embodiments of the algorithm may utilize feature-based learning techniques based on high-fidelity physics models and/or machine operation data (which would allow the algorithm to be deployed on any system) to establish a high dimensional decision boundary. As a result, detection may occur with more precision using multiple signals, making the detection more accurate with less false positives. Moreover, embodiments may detect multiple attacks on control signals, and rationalize where the root cause attack originated. For example, the algorithm may decide if a signal is anomalous because of a previous signal attack, or if it is instead independently under attack. This may be accomplished, for example, by monitoring the evolution of the features as well as by accounting for time delays between attacks.
A cyber-attack detection and localization algorithm may process a real-time cyber-physical system signal data stream and then compute features (multiple identifiers) which can then be compared to the signal-specific decision boundary. A block diagram of a system 1000 utilizing a signal-specific cyber-physical system abnormality detection and localization algorithm according to some embodiments is provided in
A real-time threat detection platform 1050 may receive the boundaries along with streams of data from the monitoring nodes. The platform 1050 may include a feature extraction on each monitoring node element 1052 and a normalcy decision 1054 with an algorithm to detect attacks in individual signals using signal specific decision boundaries, as well rationalize attacks on multiple signals, to declare which signals were attacked (or are otherwise abnormal), and which became anomalous due to a previous attack on the system via a localization module 1056. An accommodation element 1058 may generate outputs 1070, such as an anomaly decision indication (e.g., an abnormal) alert signal, a controller action, and/or a list of abnormal monitoring nodes.
During real-time detection, contiguous batches of control signal data may be processed by the platform 1050, normalized and the feature vector extracted. The location of the vector for each signal in high-dimensional feature space may then be compared to a corresponding decision boundary. If it falls within the abnormal region, then a cyber-attack may be declared. The algorithm may then make a decision about where the attack originally occurred. An attack may sometimes be on the actuators 1038 and then manifested in the sensor 1034 data. Attack assessments might be performed in a post decision module (e.g., the localization element 1056) to isolate whether the attack is related to the sensor, controller, or actuator (e.g., indicating which part of the monitoring node). This may be done by individually monitoring, overtime, the location of the feature vector with respect to the hard decision boundary. For example, when a sensor 1034 is spoofed, the attacked sensor feature vector will cross the hard decision boundary earlier than the rest of the vectors as described with respect to
According to some embodiments, it may be detected whether or not a signal is in the normal operating space (or abnormal space) through the use of localized decision boundaries and real-time computation of the specific signal features. Moreover, an algorithm may differentiate between a sensor being attacked as compared to a signal to auxiliary equipment being attacked. The control intermediary parameters and control logical(s) may also be analyzed using similar methods. Note that an algorithm may rationalize signals that become anomalous. An attack on a signal may then be identified.
A graph is provided for five data-driven features: compressor discharge temperature 1110, compressor pressure ratio 1120, compressor inlet temperature 1130, fuel flow 1140, and generator power 1150. Each graph includes a hard boundary 1112 (solid curve), inner boundary 1116 (dotted curve), and outer boundary 1114 (dashed curve) and an indication associated with current feature location for each monitoring node parameter (illustrated with an “X” on the graphs). As illustrated in
Given the example of
Note that one signal rationalization might be associated with a system time delay. That is, after a sensor is attacked there might be a period of time before the system returns to a steady state. After this delay, any signal that becomes anomalous might be due to an attack as opposed to the system responding.
The current methods for detecting abnormal conditions in monitoring nodes are limited to Fault Detection Isolation and Accommodation (“FDIA”), which itself is very limited. The hybrid cyber-attack detection and localization algorithms described herein can not only detect abnormal signals of sensors, but can also detect signals sent to auxiliary equipment, control intermediary parameters and/or control logical(s). The algorithms may also understand multiple signal attacks. One challenge with correctly identifying a cyber-attack threat is that it may occur with multiple sensors being impacted by malware. According to some embodiments, an algorithm may identify in real-time that an attack has occurred, which sensor(s) are impacted, and declare a fault response. To achieve such a result, the detailed physical response of the system must be known to create acceptable decision boundaries. This might be accomplished, for example, by constructing data sets for normal and abnormal regions by running DoE experiments on high-fidelity models. A data set for each sensor might comprise a feature vector for given threat values (e.g., temperature, airflow, etc.). Full factorial, Taguchi screening, central composite and Box-Behnken are some of the known design methodologies used to create the attack space. When models are not available, these DoE methods are also used to collect data from real-world cyber-physical systems. Experiments may be run at different combinations of simultaneous attacks. In some embodiments, the system may detect degraded/faulty operation as opposed to a cyber-attack. Such decisions might utilze a data set associated with a degraded/faulty operating space. At the end of this process, the system may create data sets such as “attack v/s normal” and “degraded v/s normal” for use while constructing decision boundaries. Further note that a decision boundary may be created for each signal using data sets in feature space. Various classification methods may be used to compute decision boundaries. For example, binary linear and non-linear supervised classifiers are examples of methods that could be used to obtain a decision boundary.
Note that multiple vector properties might be examined, and the information described with respect to
According to some embodiments, the system may localize or otherwise analyze an origin of the threat to a particular monitoring node. For example, the localizing may be performed in accordance with a time at which a decision boundary associated with one monitoring node was crossed as compared to a time at which a decision boundary associated with another monitoring node was crossed. According to some embodiments, an indication of the particular monitoring node might be included in the threat alert signal.
Some embodiments described herein may take advantage of the physics of a cyber-physical system by learning a priori from tuned high-fidelity equipment models and/or actual “on the job” data to detect single or multiple simultaneous adversarial threats to the system. Moreover, according to some embodiments, all monitoring node data may be converted to features using advanced feature-based methods, and the real-time operation of the cyber-physical system may be monitored in substantially real-time. Abnormalities may be detected by classifying the monitored data as being “normal” or disrupted (or degraded). This decision boundary may be constructed using dynamic models and may help to enable early detection of vulnerabilities (and potentially avert catastrophic failures) allowing an operator to restore the cyber-physical system to normal operation in a timely fashion.
Thus, some embodiments may provide an advanced hybrid anomaly detection algorithm to detect cyber-attacks on, for example, key cyber-physical system control sensors. The algorithm may identify which signals(s) are being attacked using control signal-specific decision boundaries and may inform a cyber-physical system to take accommodative actions. In particular, a detection and localization algorithm might detect whether a sensor, auxiliary equipment input signal, control intermediary parameter, or control logical are in a normal or anomalous state. Some examples of cyber-physical system monitoring nodes that might be analyzed include: critical control sensors; control system intermediary parameters; auxiliary equipment input signals; and/or logical commands to controller.
A cyber-attack detection and localization algorithm may process a real-time cyber-physical system signal data stream and then compute features (multiple identifiers) which can then be compared to the sensor specific decision boundary. In some embodiments, generating features may involve simply performing an identity transform. That is, the original signal might be used as it is.
Feature vectors, including those based on data-driven and watermarking values, may be generated on a monitoring-node-by-monitoring node basis and may be considered “local” to each particular monitoring node.
According to some embodiments, a watermarking-based attack detection system may have a dedicated detector (i.e., classification decision boundary) for watermarking features—or may augment the watermarking features with other features to be used in a unified global attack detector.
The monitoring node measurements 1610 are also utilized by a watermarking data collection process 1630 to create data for watermarking features extraction 1640. Feature augmentation 1650 may be performed on the extracted base (data-driven) features and watermarking features and the augmented results may be used by a unified attack detector 1670 (with appropriate decision boundaries) to create a unified system categorization (“normal” or “abnormal”).
Note that physical watermarking may utilize all system actuators or a fixed subset of actuator for watermarking injection. While the injected signals are random, the locations of injection might always be the same. As additional protection, the system may also randomize the selected injection points. The advantage of such an approach may be two-fold: (1) it may provide an additional degree of freedom in the design of the watermarking scheme and make it more difficult for an adversary to overcome, and (2) it may be leveraged to design cost-effective watermarking schemes with minimized impact on controls performance.
According to some embodiments, physical watermarking may be focused on injecting watermarking signal only into the control actuators. For some systems, set-points may also be watermarked. For example, in a wind farm, it is reasonable to assume that the set-points sent from the farm level controls to the turbines might also be watermarked.
Considering, the wind turbine actuators (e.g. blade pitch angles and generator torque) as the unit level control commands, essentially, the turbine set-points are the plant-level actuator signals commanded by the supervisory level (farm level) controller to the unit level (wind turbine) controller.
Note that the embodiments described herein may be implemented using any number of different hardware configurations. For example,
The processor 2010 also communicates with a storage device 2030. The storage device 2030 may comprise any appropriate information storage device, including combinations of magnetic storage devices (e.g., a hard disk drive), optical storage devices, mobile telephones, and/or semiconductor memory devices. The storage device 2030 stores a program 2012 and/or cyber-physical system protection engine 2014 for controlling the processor 2010. The processor 2010 performs instructions of the programs 2012, 2014, and thereby operates in accordance with any of the embodiments described herein. For example, the processor 2010 may randomly inject a watermarking signal into an injection subset of system nodes. The processor 2010 may then receive current monitoring node values over time and generate a current watermarking feature vector based on the current monitoring node values. The processor 2010 might be associated with a dedicated watermarking abnormality detection platform or a unified abnormality detection platform (e.g., that also uses data-drive feature vectors). The injection subset may be associated with a randomly selected subset of the system nodes and/or magnitudes of watermarking signals that are randomly selected.
The programs 2012, 2014 may be stored in a compressed, uncompiled and/or encrypted format. The programs 2012, 2014 may furthermore include other program elements, such as an operating system, clipboard application, a database management system, and/or device drivers used by the processor 2010 to interface with peripheral devices.
As used herein, information may be “received” by or “transmitted” to, for example: (i) the cyber-physical system protection platform 2000 from another device; or (ii) a software application or module within the cyber-physical system protection platform 2000 from another software application, module, or any other source.
In some embodiments (such as the one shown in
Referring to
The cyber-physical system identifier 2102 and description 2104 may define a particular asset or system that will be protected. The node identifier 2106 may specify a sensor, actuator, etc. of the system and indicate whether or not that node is currently being used as an injection point (and, if so, a magnitude of an injected signal as a percentage adjustment, an absolute adjustment, a random value, etc.). As illustrated in
Thus, embodiments may provide technical improvements to cyber-physical system protection. Watermarking is an active defense strategy that proactively injects additional signals into the system and collects the outcomes. The existence of the watermark is often unknown to attacker. In addition, both a watermarking signal generation scheme and injection locations may be randomized. To perform a successful spoofing and replaying attack and remain undetected, an adversary would need to acquire the following knowledge about the system:
Benefits of some embodiments described herein may include: detecting man-in-the-middle attacks or mass spoofing and replay attacks; a reduced false negative rate; and an ability to be seamlessly combined with the existing (e.g., legacy) attack/anomaly detection systems via decision fusion. Other advantages may include: increased cyber-detection coverage and to fill gaps in detecting a replay attack.; asset protection through an active defense mechanism; and increased asset reliability and an improved confidence level in the asset's integrity.
The following illustrates various additional embodiments of the invention. These do not constitute a definition of all possible embodiments, and those skilled in the art will understand that the present invention is applicable to many other embodiments. Further, although the following embodiments are briefly described for clarity, those skilled in the art will understand how to make any changes, if necessary, to the above-described apparatus and methods to accommodate these and other embodiments and applications.
Although specific hardware and data configurations have been described herein, note that any number of other configurations may be provided in accordance with embodiments of the present invention (e.g., some of the information associated with the databases described herein may be combined or stored in external systems). Moreover, although some embodiments are focused on gas turbines, any of the embodiments described herein could be applied to other types of cyber-physical systems including power grids, dams, locomotives, additive printers, data centers, airplanes, and autonomous vehicles (including automobiles, trucks, drones, submarines, etc.).
The present invention has been described in terms of several embodiments solely for the purpose of illustration. Persons skilled in the art will recognize from this description that the invention is not limited to the embodiments described but may be practiced with modifications and alterations limited only by the spirit and scope of the appended claims.
