Abstract: We disclose a system that executes an inferential model in VRAM that is embedded in a set of graphics-processing units (GPUs). The system obtains execution parameters for the inferential model specifying: a number of signals, a number of training vectors, a number of observations and a desired data precision. It also obtains one or more formulae for computing memory usage for the inferential model based on the execution parameters. Next, the system uses the one or more formulae and the execution parameters to compute an estimated memory footprint for the inferential model. The system uses the estimated memory footprint to determine a required number of GPUs to execute the inferential model, and generates code for executing the inferential model in parallel while efficiently using available memory in the required number of GPUs. Finally, the system uses the generated code to execute the inferential model in the set of GPUs.

Abstract: The disclosed embodiments relate to a system that compactly stores time-series sensor signals. During operation, the system receives original time-series signals comprising sequences of observations obtained from sensors in a monitored system. Next, the system formulizes the original time-series sensor signals to produce a set of equations, which can be used to generate synthetic time-series signals having the same correlation structure and the same stochastic properties as the original time-series signals. Finally, the system stores the formulized time-series sensor signals in place of the original time-series sensor signals.

Abstract: During operation, the system obtains the time-series sensor signals, which were gathered from sensors in a monitored system. Next, the system classifies the time-series sensor signals into stair-stepped signals and un-stair-stepped signals. The system then replaces stair-stepped values in the stair-stepped signals with interpolated values determined from un-stair-stepped values in the stair-stepped signals. Next, the system divides the time-series sensor data into a training set and an estimation set. The system then trains an inferential model on the training set, and uses the trained inferential model to replace interpolated values in the estimation set with inferential estimates. Next, the system switches roles of the training and estimation sets to produce a new training set and a new estimation set. The system then trains the inferential model on the new training set, and uses the trained inferential model to replace interpolated values in the new estimation set with inferential estimates.

Abstract: Detecting a counterfeit status of a target device by: selecting a set of frequencies that best reflect load dynamics or other information content of a reference device while undergoing a power test sequence; obtaining target electromagnetic interference (EMI) signals emitted by the target device while undergoing the same power test sequence; creating a sequence of target kiviat plots from the amplitude of the target EMI signals at each of the set of frequencies at observations over the power test sequence to form a target kiviat tube EMI fingerprint; comparing the target kiviat tube EMI fingerprint to a reference kiviat tube EMI fingerprint for the reference device undergoing the power test sequence to determine whether the target device and the reference device are of the same type; and generating a signal to indicate a counterfeit status based at least in part on the results of the comparison.

Abstract: Detecting whether a target device that includes multiple electronic components is genuine or suspected counterfeit by: performing a test sequence of energizing and de-energizing the target device and collecting electromagnetic interference (EMI) signals emitted by the target device; generating a target EMI fingerprint from the EMI signals collected; retrieving a plurality of reference EMI fingerprints from a database library, each of which corresponds to a different configuration of electronic components of a genuine device of the same make and model as the target device; iteratively comparing the target EMI fingerprint to the retrieved reference EMI fingerprints and generating a similarity metric between each compared set; and indicating that the target device (i) is genuine where the similarity metric for any individual reference EMI fingerprint satisfies a threshold test, and is a suspect counterfeit device where no similarity metric for any individual reference EMI fingerprint satisfies the test.

Abstract: The disclosed embodiments provide a system that detects sensor anomalies in a univariate time-series signal. During a surveillance mode, the system receives the univariate time-series signal from a sensor in a monitored system. Next, the system performs a staggered-sampling operation on the univariate time-series signal to produce N sub-sampled time-series signals, wherein the staggered-sampling operation allocates consecutive samples from the univariate time-series signal to the N sub-sampled time-series signals in a round-robin ordering. The system then uses a trained inferential model to generate estimated values for the N sub-sampled time-series signals based on cross-correlations with other sub-sampled time-series signals. Next, the system performs an anomaly detection operation to detect incipient sensor anomalies in the univariate time-series signal based on differences between actual values and the estimated values for the N sub-sampled time-series signals.

Abstract: Systems, methods, and other embodiments associated with autonomous cloud-node scoping for big-data machine learning use cases are described. In some example embodiments, an automated scoping tool, method, and system are presented that, for each of multiple combinations of parameter values, (i) set a combination of parameter values describing a usage scenario, (ii) execute a machine learning application according to the combination of parameter values on a target cloud environment, and (iii) measure the computational cost for the execution of the machine learning application. A recommendation regarding configuration of central processing unit(s), graphics processing unit(s), and memory for the target cloud environment to execute the machine learning application is generated based on the measured computational costs.

Abstract: Systems and methods are described that estimates a remaining useful life (RUL) of an electronic device. Time-series signals gathered from sensors in the electronic device are received. Statistical changes are detected in the set of time-series signals that are deemed as anomalous signal patterns. Anomaly alarms are generated, wherein an anomaly alarm is generated for each of the anomalous signal patterns. An irrelevance filter is applied to the set of anomaly alarms to produce filtered anomaly alarms, wherein the irrelevance filter removes anomaly alarms associated with anomalous signal patterns that are not correlated with previous failures of similar electronic devices. A logistic-regression model is used to compute an RUL-based risk index for the electronic device based on the filtered anomaly alarms. When the risk index exceeds a risk-index threshold, a notification is generated indicating that the electronic device has a limited remaining useful life.

Abstract: Systems, methods, and other embodiments associated with off-duty-cycle-robust machine learning for anomaly detection in assets with random downtimes are described. In one embodiment, a method includes inferring ranges of asset downtime from spikes in a numerical derivative of a time series signal for an asset; extracting an asset downtime signal from the time series signal based on the inferred ranges of asset downtime; determining that the asset downtime signal carries telemetry based on the variance of the asset downtime signal; training a first machine learning model for the asset downtime signal; detecting a first spike in the numerical derivative of the time signal that indicates a transition to asset downtime; and in response to detection of the first spike, monitoring the time series signal for anomalous activity with the trained first machine learning model.

Abstract: The disclosed embodiments relate to a system that camouflages EMI fingerprints in EMI emissions from a computing system to enhance system security. During operation, the system monitors the EMI emissions from the computer system during operation of the computer system to produce corresponding EMI signals. Next, the system determines a dynamic amplitude of the EMI emissions based on the EMI signals. If the dynamic amplitude of the EMI emissions drops below a threshold value, the system executes synthetic transactions, which have interarrival times that, when superimposed on a workload of the computer system, cause the computer system to produce randomized EMI emissions.

Abstract: Techniques for providing decision rationales for machine-learning guided processes are described herein. In some embodiments, the techniques described herein include processing queries for an explanation of an outcome of a set of one or more decisions guided by one or more machine-learning processes with supervision by at least one human operator. Responsive to receiving the query, a system determines, based on a set of one or more rationale data structures, whether the outcome was caused by human operator error or the one or more machine-learning processes. The system then generates a query response indicating whether the outcome was caused by the human operator error or the one or more machine-learning processes.

Abstract: The system receives exemplary time-series sensor signals comprising ground truth versions of signals generated by a monitored system associated with a target use case and a synchronization objective, which specifies a desired tradeoff between synchronization compute cost and synchronization accuracy for the target use case. The system performance-tests multiple synchronization techniques by introducing randomized lag times into the exemplary time-series sensor signals to produce time-shifted time-series sensor signals, and then uses each of the multiple synchronization techniques to synchronize the time-shifted time-series sensor signals across a range of different numbers of time-series sensor signals, and a range of different numbers of observations for each time-series sensor signal. The system uses the synchronization objective to evaluate results of the performance-testing in terms of compute cost and synchronization accuracy.

Abstract: The disclosed embodiments relate to a system that facilitates development of machine-learning techniques to perform prognostic-surveillance operations on time-series data from a monitored system, such as a power plant and associated power-distribution system. During operation, the system receives original time-series signals comprising sequences of observations obtained from sensors in the monitored system. Next, the system decomposes the original time-series signals into deterministic and stochastic components. The system then uses the deterministic and stochastic components to produce synthetic time-series signals, which are statistically indistinguishable from the original time-series signals. Finally, the system enables a developer to use the synthetic time-series signals to develop machine-learning (ML) techniques to perform prognostic-surveillance operations on subsequently received time-series signals from the monitored system.

Abstract: Systems, methods, and other embodiments associated with automated calibration in electromagnetic scanners are described. In one embodiment, a method includes: detecting one or more peak frequency bands in electromagnetic signals collected by the electromagnetic scanner at a geographic location; comparing the one or more peak frequency bands to broadcast frequencies assigned to local radio stations of the geographic location; and indicating that the electromagnetic scanner is calibrated by finding at least one match between one peak frequency band of the peak frequency bands and one of the broadcast frequencies. An electromagnetic scanner may be recalibrated based on comparing the one or more peak frequency bands to broadcast frequencies.

Abstract: Systems, methods, and other embodiments associated with autonomous cloud-node scoping for big-data machine learning use cases are described. In some example embodiments, an automated scoping tool, method, and system are presented that, for each of multiple combinations of parameter values, (i) set a combination of parameter values describing a usage scenario, (ii) execute a machine learning application according to the combination of parameter values on a target cloud environment, and (iii) measure the computational cost for the execution of the machine learning application. A recommendation regarding configuration of central processing unit(s), graphics processing unit(s), and memory for the target cloud environment to execute the machine learning application is generated based on the measured computational costs.

Abstract: The disclosed embodiments provide a system that estimates a remaining useful life (RUL) for a fan. During operation, the system receives telemetry data associated with the fan during operation of the critical asset, wherein the telemetry data includes a fan-speed signal. Next, the system uses the telemetry data to construct a historical fan-speed profile, which indicates a cumulative time that the fan has operated in specific ranges of fan speeds. The system then computes an RUL for the fan based on the historical fan-speed profile and empirical time-to-failure (TTF) data, which indicates a TTF for the same type of fan as a function of fan speed. Finally, when the RUL falls below a threshold, the system generates a notification indicating that the fan needs to be replaced.

Abstract: During operation, the system receives time-series signals gathered from sensors in a utility system asset. Next, the system uses an inferential model to generate estimated values for the time-series signals, and performs a pairwise differencing operation between actual values and the estimated values for the time-series signals to produce residuals. The system then performs a sequential probability ratio test (SPRT) on the residuals to produce SPRT alarms. Next, the system applies an irrelevance filter to the SPRT alarms to produce filtered SPRT alarms, wherein the irrelevance filter removes SPRT alarms for signals that are uncorrelated with previous failures of similar utility system assets. The system then uses a logistic-regression model to compute an RUL-based risk index for the utility system asset based on the filtered SPRT alarms. When the risk index exceeds a threshold, the system generates a notification indicating that the utility system asset needs to be replaced.

Abstract: A double-blind comparison is performed between prognostic-surveillance systems, which are located on a local system and a remote system. During operation, the local system inserts random faults into a dataset to produce a locally seeded dataset, wherein the random faults are inserted into random signals at random times with variable fault signatures. Next, the local system exchanges the locally seeded dataset with a remote system, and in return receives a remotely seeded dataset, which was produced by the remote system by inserting different random faults into the same dataset. Next, the local system uses a local prognostic-surveillance system to analyze the remotely seeded dataset to produce locally detected faults. Finally, the local system determines a performance of the local prognostic-surveillance system by comparing the locally detected faults against actual faults in the remotely seeded fault information. The remote system similarly determines a performance of a remote prognostic-surveillance system.

Abstract: During operation, the system uses N sensors to sample an electromagnetic interference (EMI) signal emitted by a target asset while the target asset is running a periodic workload, wherein each of the N sensors has a sensor sampling frequency f, and wherein the N sensors perform sampling operations in a round-robin ordering with phase offsets between successive samples. During the sampling operations, the system performs phase adjustments among the N sensors to maximize phase offsets between successive sensors in the round-robin ordering. Next, the system combines samples obtained through the N sensors to produce a target EMI signal having an EMI signal sampling frequency F=f×N. The system then generates a target EMI fingerprint from the target EMI signal. Finally, the system compares the target EMI fingerprint against a reference EMI fingerprint for the target asset to determine whether the target asset contains any unwanted electronic components.

Abstract: The disclosed embodiments relate to a system that characterizes susceptibility of an inferential model to follow signal degradation. During operation, the system receives a set of time-series signals associated with sensors in a monitored system during normal fault-free operation. Next, the system trains the inferential model using the set of time-series signals. The system then characterizes susceptibility of the inferential model to follow signal degradation. During this process, the system adds degradation to a signal in the set of time-series signals to produce a degraded signal. Next, the system uses the inferential model to perform prognostic-surveillance operations on the set of time-series signals with the degraded signal. Finally, the system characterizes susceptibility of the inferential model to follow degradation in the signal based on results of the prognostic-surveillance operations.