Intelligent hardware monitoring and fault diagnosis using sensor data analysis

Abstract

The invention discloses a test and measurement instrument. Comprising one or more ports connected to hardware under test (HUT), a set of sensors connected to the HUT and an instrument, a display to display one or more signal representations from at least one of the HUT and from the one or more sensors of the set of sensors, and one or more processors configured to execute code to cause the one or more processors to acquire data from the set of sensors; forming one or more data sets according to data acquired from the sensor group; applying one or more machine learning models to the one or more data sets; and receive a predictive analysis regarding the HUT from the one or more machine learning models.

Description

[0001] Cross-reference with related applications

[0002] Pursuant to 35 U.S.C., 119, this disclosure claims priority to Indian Provisional Patent Application No. 20242052998, filed July 11, 2024, entitled “Smart Motor Monitoring and Fault Diagnosis Using Sensor Data Analysis,” the entire contents of which are incorporated herein by reference. Technical Field

[0003] This invention relates to the testing of motors and hardware with moving parts, and more specifically, to the application of machine learning to predict future hardware failures. Background Technology

[0004] Typically, designers use IMDA (Inverter Motor Drive Analysis) software to test motors during the design and commissioning phases of the motor design workflow. This software uses voltage and current waveforms. It also typically measures power quality, DQ0 (Direct Quadrature Zero), and harmonics.

[0005] One common method for diagnosing motors (such as AC motors, brushless DC motors, and single-phase or three-phase motors) involves using motor current characteristic analysis (MCSA) to detect motor faults. A major drawback of this type of motor analysis is that these motors cannot be analyzed during internal circuit operation, and these tests take place in controlled environments or laboratory conditions. Motor designers typically do not test motors while they are in operation.

[0006] Electric motors face various operational and environmental hazards that can generate thermal stress on the windings, reducing the motor's ability to insulate components from thermal stress and impacting its expected lifespan. Therefore, motor monitoring has become a crucial aspect and the most effective strategy for locating potential problems. Until recently, preventing motor failures was complex and expensive. However, with declining costs of meters and sensors, and advancements in big data technologies, motor monitoring has become accurate and affordable. All of these factors are expected to drive growth in the global motor monitoring market during the forecast period.

[0007] According to https: / / www.mordorintelligence.com / industry-reports / motor-monitoring-market, the motor monitoring market size was estimated at US$2.67 billion in 2024 and is projected to reach US$4.4 billion by 2029, growing at a CAGR of 10.46% during the forecast period (2024-2029). The table below summarizes the survey findings.

[0008] Research period 2019-2029 Market size (2024) $2.67 billion Market size (2029) $4.4 billion Compound annual growth rate (CAGR) 10.46％ Fastest growing market Asia Pacific Largest market North America.

[0009] The main players in this market are ABB, National Instruments, Honeywell, Siemens, and General Electric. These motors consume 70% of the industry's electricity, demonstrating their importance. The market is expanding as consumer preferences shift towards a "left-hand" approach and a seamless user experience.

[0010] As such a growing market enables companies that provide test and measurement solutions, they too will experience growth. Attached Figure Description

[0011] Figure 1 A flowchart illustrating the motor design workflow is shown.

[0012] Figure 2 An example of an experimental setup for testing the hardware under test (HUT) is shown.

[0013] Figure 3 An embodiment of a sensor mounted on an AC motor is shown.

[0014] Figure 4 A flowchart illustrating an embodiment of a method for training and selecting machine learning models is shown.

[0015] Figure 5 A flowchart illustrating an embodiment of feature extraction and modeling is shown.

[0016] Figure 6 Examples of vibration signals captured from an AC motor in the time and frequency domains are shown.

[0017] Figure 7 An example of a user interface for measuring configuration settings is shown.

[0018] Figure 8 An example of a user interface for configuring machine learning models is shown.

[0019] Figure 9 A flowchart illustrating an example of using machine learning to predict hardware failures is shown.

[0020] Figures 10A-10B The correlation matrices between features and outputs, as well as between features, are shown.

[0021] Figure 11A-11B Histograms of correlation and confusion matrices are shown.

[0022] Figure 12 An example of the data acquisition trend of the vibration graph is shown.

[0023] Figure 13 An example of a diagram showing the centerline of the vibrating shaft is shown.

[0024] Figure 14 An example of a prediction graph is shown. Detailed Implementation

[0025] The embodiments described herein effectively address the problems of predictive maintenance and fault diagnosis, which in turn reduces the uptime of the hardware under test (HUT). HUTs can include a wide variety of hardware types, including AC motors integrated with electric vehicles (EVs) and industrial applications. The embodiments described herein apply to all types of motors, including brushless DC motors, single-phase and three-phase motors, but are not limited to motors. Other types of HUTs may relate to devices with moving parts, whether these parts move under the power of a motor or under the power of another source. For example, the embodiments described herein apply to the high-speed shaft of a wind turbine that moves in response to wind pressure. As used herein, the term HUT includes these examples and more.

[0026] Real-time sensors on HUTs allow for the collection of vast amounts of data. Machine learning applications can use this data to accurately predict future failures. In test and measurement systems, artificial intelligence (AI) and machine learning (ML) have become powerful tools, revolutionizing data analysis and the extraction of insights from it. Machine learning modeling leveraging AI has become a crucial technique in data processing, providing valuable tools for monitoring live / running HUT systems under dynamic loads. For ease of discussion, the term machine learning used here also encompasses various aspects of generative artificial intelligence.

[0027] The embodiments described here relate to sensor-based HUT fault analysis, where sensors monitor various fault causes such as transient voltages (e.g., surges, voltage and current imbalances, vibration, high operating temperatures, HUT overload, and misalignment). Machine learning models can be used for predictive maintenance of the HUT and to predict faults within the HUT. While the discussion below focuses on motors, as mentioned above, the embodiments are applicable to other types of systems operating at other power levels. Hardware designers will typically be the users in this process, as the analysis being performed impacts the design of the HUT. The term "user" used herein applies to both the designers who design the hardware and the users who monitor HUT operation for maintenance and repair.

[0028] Various industrial and automotive sensors collect analog data to understand and predict the future health of systems. Using artificial intelligence to collect and analyze data from these sensors plays a crucial role in HUT monitoring challenges. For example, by analyzing vibration data from sensors mounted on the HUT, this analysis can provide insights for predictive maintenance. Machine learning can detect anomalies and predict potential failures before they escalate. This proactive approach minimizes downtime and reduces maintenance costs. Machine learning algorithms can identify patterns in vibration signals to allow for the diagnosis of problems such as broken rotor bars, bearing defects, and misalignment. Early detection of these problems allows for timely corrective action. Machine learning leverages historical data to learn HUT behavior, providing insights into optimal operating conditions and failure trends. Machine learning algorithms can run on embedded hardware, such as field-programmable gate arrays, for real-time monitoring, enabling continuous assessment of the HUT's health. In general, machine learning and artificial intelligence enhance HUT monitoring by predicting failures, optimizing maintenance, and ensuring reliable operation.

[0029] As discussed above, it is important to test HUT on an actual test bench. Figure 1 An example of the HUT design workflow is shown. At 10, the process begins by defining the application-specific requirements of the HUT. At 12, the various components used by the HUT are characterized. As discussed above, at 14, the user (in this case, the designer) simulates and optimizes the HUT. The process section within box 22 includes field testing, allowing designers to finalize their design and deploy and install the HUT. Electrical and mechanical testing is performed at 16, the design is finalized at 18, and deployment occurs at 20. The process section within box 22 benefits from the application of machine learning.

[0030] As an example of a HUT (Host-to-Under-Load), key parameters such as vibration, voltage, current, temperature, acceleration, and stress can be captured from sensor outputs on a running motor under load. By combining machine learning modeling and AI prediction of future failures, the health status of the HUT can be dynamically visualized and analyzed. This presents a new opportunity for the test and measurement industry. The examples here are applicable to multiple industries to improve HUT operation. These improvements include, but are not limited to, efficient and fail-safe industrial drives such as conveyor drives and pumps, electric mobility from cars to drones and electric aircraft, point-of-load converters in data centers, safety robots and mobile medical robots with longer battery life, and efficient and predictable photovoltaic inverters.

[0031] As discussed in further detail, users can save these parameters as library files for other HUTs with similar specifications, thereby establishing internal knowledge of the HUT's functionality. An AI assistant, software employing machine learning, can perform a series of operations on the instrument that help predict future failures and improve HUT performance through mechanical parameter analysis. The AI ​​assistant provides interface options for the AI ​​model that performs predictive analysis of the HUT.

[0032] These embodiments facilitate... Figure 1 HUT testing is performed in box 16 of the workflow. The machine learning model adapts based on instrument usage during debugging / testing. Users do not need any large datasets to build the model. The datasets here typically include much smaller datasets than commonly used datasets, as the models learn dynamically from streaming sensor inputs during their training. Once integrated, users can use test and measurement instruments such as oscilloscopes to test the mechanical / environmental parameters of the running HUT. In some embodiments, the test and measurement instruments capture signals from the HUT sensors for vibration analysis, which helps identify HUT component harmonics. In some embodiments, the test and measurement instruments can act as powerful digital converters of sensor data. Error waveforms are saved for machine learning modeling and artificial intelligence prediction.

[0033] Various types of sensors, including vibration sensors, play a crucial role in identifying potential faults before they occur. Vibration sensors can take many forms, including MEMS (Micro-Electro-Mechanical Systems) sensors. These sensors are reliable and cost-effective, especially performing well at low frequencies. Low-frequency monitoring makes MEMS sensors suitable for use in slowly rotating machinery. Piezoelectric vibration sensors convert mechanical vibrations into electrical signals. A wide variety of Internet of Things (IoT) industrial sensors are available from a diverse range of suppliers.

[0034] Creating a machine learning model based on sensor data collected from running hardware (such as HUT) involves several steps. The process aims to analyze data from test and measurement instruments such as oscilloscopes. This process involves integrating hardware interfaces, data collection, data preprocessing, model development, visualization, and analysis.

[0035] Figure 2 An embodiment of the test environment is illustrated. Initially, sensors like sensor 30 need to be mounted on the HUT 31. The sensors must be securely fixed, and their position must be such that they capture the operating characteristics of the HUT, as shown by sensor 30. The mounted sensor 30 unit provides sensor data that the system 35 can collect in many different ways. For the HUT 31, a triaxial sensor, rather than a single-axis or radial sensor, can provide more information, thus enabling better fault detection. Figure 3The configuration of the triaxial sensors is shown, including the orientation of the axial sensor 44, radial sensor 42, and tangential sensor 46. The sensors (such as sensor 30) are adjusted to allow for sensor calibration and detection of sensor malfunctions. The test settings differ, and sensor 30 can capture data in the event of a sensor malfunction.

[0036] A data logger 34, such as a dedicated data acquisition system or microcontroller, can collect data. The data logger 34 can then transfer the data to some form of storage 33, such as an SD card, USB drive, etc. Test and measurement instruments 32 can also collect data directly. Most of the data will likely include analog data, and many test and measurement instruments, such as oscilloscopes, may include analog-to-digital converters to convert the data. Test and measurement instruments 32 or data logger 34 can also collect data or transfer it to remote storage 40, such as cloud-based storage. Remote storage 40 can also receive datasets from sensors 30 and results from other components to build the aforementioned library.

[0037] Test and measurement instrument 32 may also have probes (not shown) that can be directly or connected to HUT 31 using test fixtures. Test and measurement instrument 32 may also include one or more processors for analyzing data and receiving and acting on user input. The presence of one or more processors and internal memory in test and measurement instrument 32 also allows machine learning models and artificial intelligence analysis 36 to run on test and measurement instrument 32. As an alternative to or complement to test and measurement instrument 32 containing machine learning models and artificial intelligence analysis, computing device 38 (such as a personal computer) may contain machine learning models and artificial intelligence analysis. The various tasks and components discussed below may be distributed among test and measurement instrument 32, computing device 38, and devices connected via cloud 40.

[0038] Figure 4 The overall process for collecting and analyzing data gathered from sensors is illustrated. Figure 4 At position 50, the HUT 31 operates under normal load conditions and any specific conditions of interest, such as various loads or speeds. Sensor 30 provides data that the system can convert into one or more datasets representing the operating range of the HUT 31. As described above, the test and measurement instrument 32 can act as a real-time monitor, or data can be recorded and stored locally or remotely. Using the test and measurement instrument 32 as a real-time monitor has advantages because the user can view the time-domain waveform of the running HUT 31. Probes and / or sensors can be connected to the test and measurement instrument 32 via the cloud, antenna, or directly.

[0039] After or during data collection, system 35 (i.e., a combination of test and measurement instruments 32 and / or computing devices 38 and remote storage 40) merges the data into one or more datasets at point 52, each dataset requiring preprocessing to normalize the data. One or more datasets are normalized to eliminate bias, improve feature extraction, and refine the training process. For example, a normalization process is applied to eliminate any bias from the collected datasets:

[0040]

[0041] Feature extraction process 54 extracts features from data related to the operational characteristics of HUT 31. (About...) Figure 5 The feature extraction process is discussed in more detail. Then, at point 56, the machine learning model is applied to the dataset. The model applied to the dataset at point 45 could be... Figure 2 The machine learning model 36 or 38, depending on where it is run, can include one of many different types of predictive models, including random forests, decision trees, etc., and can be used for predictive maintenance of HUT 31 and to predict possible failures. Users can select a model via user input on test and measurement instrument 32 or computing device 38, or users can select "AUTO" on the user interface shown later. The AUTO option involves selecting the appropriate regression model with the smallest error for the specific environment in which HUT 31 operates. Model options include regression models such as linear, multivariate, multinomial, decision tree, neural network (several types), or clustering models, as examples but not limited to.

[0042] This process can apply different error measures to determine the most suitable model. Two of these are mean squared error (MSE) and root mean square error (RMSE). The minimum or lowest error is the criterion for selecting a suitable model. MSE is calculated as follows:

[0043]

[0044] RMSE is calculated as follows:

[0045]

[0046] Where n is the number of data points, Y i It is the actual i-th measurement value, and The observed value y in equation (2) and equation (3) i and The corresponding predicted value. The AUTO method = min(MSE(regression model, decision tree, neural network, clustering model). The output will be displayed as scalar results and trend points.

[0047] Back Figure 4 At point 58, the collected dataset is used to train the model. This model is trained to identify patterns or predict outcomes based on sensor data. This process may occur dynamically as the model receives data streams from sensors. Then, at point 60, the resulting trained model makes predictions, which may result in the HUT being determined to be healthy or having incorrect HUT parameters. Incorrect HUT parameters may lead to maintenance being required, or the HUT being considered faulty. The model's results can include predictive analytics such as failure time, predicted maintenance needs, etc. This process uses a test dataset to evaluate the model's performance. Model parameters can be tuned as needed to improve accuracy and reduce overfitting.

[0048] Figure 5 An example of the feature extraction and model evaluation process is illustrated. As described above, sensor data is converted into a training dataset. Data preprocessing, such as the normalization / standardization process described above, is performed in the top process at 70. Features are extracted from the data at 72, and the feature matrix used to train the model is obtained at 74. The top process in the figure uses unsupervised learning, where the model is essentially self-trained. The bottom process employs classification, where the data is preprocessed at 76 and feature extraction is performed at 78. However, in the classification scheme, feature extraction produces feature vectors. These feature vectors are then used to predict the performance of the HUT. These two processes can be used in series as mutual quality control.

[0049] In a self-supervised learning (SSL) setup for HUT fault detection, the model learns meaningful representations (embeddings) from raw sensor data without requiring labeled examples. These embeddings capture patterns in the HUT's behavior and serve as rich features for downstream tasks. To predict the time to failure, a regression model is trained on these embeddings using some labeled fault data to estimate the HUT's remaining lifetime (RUL). This approach enables early fault detection and real-time fault prediction even with limited labeled data. Fault prediction also considers the movement / drift of runtime state data towards fault / abnormal patterns.

[0050] When using sensor outputs to model the behavior or condition of a running HUT, from such Figure 5 72 or 78 in Figure 4 Extracting the correct features from the 54 data points is crucial for developing accurate and reliable machine learning models.

[0051] "Features" can include almost anything extracted from the raw data. It can be the raw data itself, a combination of data from different sensors, a statistical analysis of several measurements such as mean, variance, standard deviation, etc., or a transformation such as a Fourier transform to the frequency domain.

[0052] Users can select features and then use these features to train a model. From then on, whenever a process uses the model for prediction / classification, the process must extract the same features from new data. As an example, but not limited to, features can include one or more of the following: time-domain measurements, frequency-domain parameters, harmonic analysis such as by Fast Fourier Transform (FFT), temperature rise (where a gradual increase in temperature can indicate friction, misalignment, or electrical problems), and vibration modes (where changes in vibration modes can indicate imbalance, bearing failure, or misalignment). Vibration analysis is often one of the most critical features for designers. Vibration analysis can detect a variety of faults at an early stage, including imbalance, bearing failure, and gear failure. These are just some examples of many other possibilities. Combining vibration data with other sensor outputs such as temperature, current consumption, and acoustic emissions provides a more comprehensive understanding of the HUT's condition. The integration of these data sources, along with sophisticated machine learning models, can significantly enhance predictive maintenance strategies, reduce unplanned downtime, and extend the HUT's lifespan.

[0053] Using frequency domain parameters allows for a clearer understanding of complex vibration waveforms that are sometimes difficult to understand in the time domain. Test and measurement instruments can convert model outputs into a format that users can visualize on the test and measurement instruments. This conversion may involve generating time series predictions or simulating sensor outputs under assumed conditions. Sensor data can be displayed as voltages in volts per second in the time domain and its accompanying frequency domain plot. Time-domain sensor information can be converted to frequency-domain sensor information as the dominant peak value = max(ABS(FFT(sensor_output_data)) in equation (4). Figure 6 Examples of test and measurement instruments, such as oscilloscopes, are shown, displaying the model's output in real time. Users can use the analysis tools on these test and measurement instruments to gain deeper insights into the HUT or HUT operation and the model's predictive capabilities. The model can then be tuned as needed based on these observations.

[0054] Figure 7 The user interface for configuring measurements is shown. Figure 7 The user interface 82 on the left shows the options available in one embodiment of the interface. The user selects the type of graph to use from the measurement. In this embodiment of user interface 82, supported graphs may include vibration trend, characteristic trend, predicted trend, axis centerline graph, correlation / confusion histogram, and correlation / confusion table. Figure 7 The user interface 84 on the right illustrates an example of sensor configuration. In this particular embodiment, the user configures the properties of the axis sensor used in the HUT analysis, such as... Figure 3 Those shown.

[0055] Figure 8The user interface 86 is shown when a user selects a configuration for AI-machine learning modeling. In this example, the user selects the time parameter and whether to use real-time data or data from a file. Typically, selecting AI-machine learning modeling on test and measurement instruments causes the proposed application to begin collecting data on key features, as well as metadata including vibration, temperature, etc. Before extracting key features, the acquired waveforms and data are preprocessed and cleaned, such as resampling, detrending, smoothing, filtering, etc. The extracted features are used as the training set to build the model. Once a sufficient training set has been collected, the model is dynamically trained in the background. The model is cross-validated against another dataset to verify accuracy and correctness. If the model's accuracy is high enough, the application allows the user to use the model to predict results as well as measurement results. This process is repeated or performed simultaneously for each available model.

[0056] Figure 9 A flowchart illustrating an overall embodiment of a machine learning process for predicting HUT or HUT behavior is shown. The process begins at 90 with the sensor installed on the HUT. The HUT is then started, and operational data is collected at 92. Then, as discussed above, at 94, the data is transmitted directly, via a data logger, or from the cloud to test and measurement instruments, and at 96, the data may also be sent to storage, such as the cloud. At 98, the data undergoes preprocessing. At 100, the model is selected. As mentioned above, the user can select the model, or the system can select the model when the user selects AUTO. At 102, models and associated data for specific HUTs and environments can be stored to build a HUT library. At 104, the selected model produces predictive results indicating predictive maintenance or health / failure HUT indicators. Then, at 106, the model is updated using the sensor data used in this cycle to adjust its operation as needed.

[0057] As mentioned above, an analysis of the model involves correlation and confusion matrices. The correlation matrix displays the interdependencies of extracted features based on scalar values. The correlation matrix helps to understand the relationships between features and identify potential problems such as multicollinearity. Figure 10A An average measurement table is shown, indicating the relationship between failures and malfunctions and bearing wear, misalignment, and imbalance. Figure 10B It shows that both horizontal and vertical vibrations are directly related to velocity, temperature, and armature current. Figure 10A and 10B The shaded boxes in the diagram indicate possible causes of the malfunction. For example, if the correlation between vibration and temperature is high, close to 1, it means that vibration will increase at higher temperatures. Figure 10B In this example, horizontal and vertical vibrations are directly related to inertia.

[0058] Correlation can also be viewed in bar charts, which can help visualize the interdependence of feature distributions. Figure 11A Histograms of correlation and confusion are shown. Figure 11B Examples of 3D histograms of confusion matrices are shown. The Y-axis in these plots includes a correlation measure.

[0059] use Figure 11B The confusion histogram can be used to develop a confusion matrix, which typically compares the true negatives, true positives, false negatives, and false positives between predicted and actual results to show how frequently the model is confused between different categories.

[0060] Prediction normal Prediction error Truly normal 90 (true positive) 10 (False Positives) Real error 5 (False negative) 95 (True negative)

[0061] In addition, the data acquisition trend chart visualizes the changes of measurement parameters over time based on the order of data collection. Figure 12 An ACQ trend plot of the extracted features from the vibration data is shown. This plot includes user-defined limit lines / masks. This helps distinguish between good and bad HUT behavior.

[0062] Figure 13 A centerline diagram is shown for analyzing the rotor shaft position relative to the bearings. The centerline diagram can reveal important issues, including imbalance, bearing stiffness, and fluid instability. This diagram is constructed from X and Y position sensor data over time. The centerline diagram is an incremental plot, and the crosshairs indicate the position of the running shaft as viewed from a cross-sectional view. This diagram helps to understand the mechanical problems that accumulate in the rotor over time as the shaft position moves. The X and Y axes indicate the outputs of the two position sensors. The circle indicates the diameter of the bearing clearance. The diameter of this circle is the maximum value that the rotor shaft can move in the horizontal or vertical direction. This diameter is an input from the user regarding the plotting settings.

[0063] Figure 14 An example of a pass / fail graph is shown. The relationship between vibration values ​​and time shows that the HUT failed after 8 hours of operation. Limits can be defined for scalar results, and masks can be inserted for pass / fail. The above graph provides an example of a dynamic plot indicating the model results.

[0064] In this way, sensors attached to the HUT can provide data to a machine learning model to determine the predicted behavior of the HUT, such as pass / fail, health or failure, or the need for maintenance. The machine learning model is used for predictive analysis of motor performance. Several machine learning models may be available, and the test and measurement instruments or computing devices operating them can select the model with the least error. These models can be supervised learning models and unsupervised learning models, including self-supervised learning models.

[0065] The aspects of this disclosure can operate on specially created hardware, firmware, a digital signal processor, or a specially programmed general-purpose computer including a processor that operates according to programmed instructions. The terms controller or processor used herein are intended to include microprocessors, microcomputers, application-specific integrated circuits (ASICs), and special-purpose hardware controllers. One or more aspects of this disclosure can be embodied in computer-usable data and computer-executable instructions, for example, in one or more program modules executed by one or more computers (including monitoring modules) or other devices. Typically, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types when executed by a processor in a computer or other device. Computer-executable instructions can be stored on a non-transitory computer-readable medium, such as a hard disk, optical disk, removable storage medium, solid-state memory, random access memory (RAM), etc. As those skilled in the art will understand, the functionality of a program module can be combined or distributed in various aspects as needed. Furthermore, this functionality can be wholly or partially embodied in firmware or hardware equivalents, such as integrated circuits, FPGAs, etc. Specific data structures can be used to more efficiently implement one or more aspects of this disclosure, and these data structures are contemplated within the scope of the computer-executable instructions and computer-usable data described herein.

[0066] In some cases, the disclosed aspects may be implemented in hardware, firmware, software, or any combination thereof. The disclosed aspects may also be implemented as instructions carried on or stored on one or more non-transitory computer-readable media, which may be read and executed by one or more processors. Such instructions may be referred to as a computer program product. As described herein, a computer-readable medium means any medium accessible by a computing device. By way of example and not limitation, a computer-readable medium may include computer storage media and communication media.

[0067] Computer storage media refers to any medium that can be used to store computer-readable information. By way of example and not limitation, computer storage media may include RAM, ROM, electrically erasable programmable read-only memory (EEPROM), flash memory or other storage technologies, optical disc read-only memory (CD-ROM), digital video disc (DVD) or other optical disc storage, magnetic tape cassettes, magnetic tape, disk storage or other magnetic storage devices, and any other volatile or non-volatile, removable or non-removable medium implemented in any technology. Computer storage media does not include the signal itself or the transient form of signal transmission.

[0068] A communication medium is any medium that can be used for computer-readable information communication. By way of example and not limitation, a communication medium may include coaxial cable, fiber optic cable, air, or any other medium suitable for communication of electrical, optical, radio frequency (RF), infrared, acoustic, or other types of signals.

[0069] Example [To be filled in after the weights are finalized]

[0070] Illustrative examples of the disclosed techniques are provided below. An embodiment of these techniques may include one or more examples described below, as well as any combination of these examples.

[0071] Example 1 is a test and measurement instrument comprising: one or more ports connected to a hardware under test (HUT); a set of sensors connected to the HUT and the instrument; a display for displaying one or more signal representations from at least one of the HUTs and from one or more sensors in the set of sensors; and one or more processors configured to execute code to cause the one or more processors to: acquire data from the set of sensors; form one or more datasets based on the data acquired from the set of sensors; apply one or more machine learning models to the one or more datasets; and receive predictive analytics about the HUT from the one or more machine learning models.

[0072] Example 2 is a test and measurement instrument of Example 1, wherein a group of sensors connected to the instrument is connected to the instrument via a sensor data collector that includes either a microcontroller or a data logger.

[0073] Example 3 is a test and measurement instrument of either Example 1 or 2, wherein the code that causes one or more processors to acquire data includes code that causes one or more processors to acquire data from one of the sensors or from a remote memory.

[0074] Example 4 is a test and measurement instrument of any one of Examples 1 to 3, wherein the code that causes one or more processors to form one or more datasets includes code that causes one or more processors to normalize the data and form one or more training datasets, one or more test datasets, and one or more prediction datasets.

[0075] Example 5 is a test and measurement instrument of any of Examples 1 through 4, wherein the code that causes one or more processors to apply one or more machine learning models to one or more datasets includes code that causes one or more processors to perform the following operations: receive input through a user interface that identifies a selected model to be applied to one or more datasets; and apply the selected model to one or more datasets.

[0076] Example 6 is a test and measurement instrument of Example 5, wherein identifying the input of the selected model includes automatic selection, and the code for applying machine learning to one or more processors includes code for applying a machine learning model with the lowest error rate to one or more processors.

[0077] Example 7 is a test and measurement instrument of any of Examples 1 through 6, wherein one or more processors are further configured to execute code that enables one or more processors to dynamically train one or more machine learning models at runtime using data streaming from sensors.

[0078] Example 8 is a test and measurement instrument of any of Examples 1 through 7, wherein the code that causes one or more processors to apply machine learning to a dataset includes code that causes one or more processors to perform the following operations: apply one or more machine learning models to one or more training datasets to train one or more machine learning models; use one or more test datasets to test one or more machine learning models; and tune the parameters of multiple models to improve the accuracy of multiple models.

[0079] Example 9 is a test and measurement instrument of Example 8, wherein the code that causes one or more processors to apply one or more machine learning models to one or more datasets to train one or more machine learning models includes code that causes one or more processors to perform feature extraction on one or more datasets of sensor data.

[0080] Example 10 is a test and measurement system of any of Examples 1 through 9, wherein one or more processors are further configured to execute code to cause the one or more processors to produce a graph indicating the results from one or more machine learning models.

[0081] Example 11 is a method comprising: receiving data from a set of sensors connected to the hardware under test (HUT); forming one or more datasets based on the data acquired from the set of sensors; applying one or more machine learning models to the one or more datasets; and receiving one or more predictions from the one or more machine learning models regarding the probability of failure of the HUT.

[0082] Example 12 is a method of Example 11, wherein receiving data includes receiving data from one of the sensors or from a remote memory.

[0083] Example 13 is a method of either Example 11 or 12, wherein forming one or more datasets includes normalizing the data and forming one or more training datasets, one or more test datasets, and one or more prediction datasets.

[0084] Example 14 is a method of any of Examples 11 to 13, applying one or more machine learning models to one or more datasets, including: receiving input through a user interface to identify the selected model to be applied to one or more datasets; and applying the selected model to one or more datasets.

[0085] Example 15 is a method of any of Examples 11 to 13, wherein identifying the input to the selected model includes automatic selection, and applying the machine learning model includes applying the machine learning model with the lowest error rate.

[0086] Example 16 is a method of any of Examples 11 through 13, and also includes one or more machine learning models that are dynamically trained at runtime using data streamed from sensors.

[0087] Example 17 is a method of any of Examples 11 through 13, wherein applying machine learning to a dataset includes: applying one or more machine learning models to one or more training datasets to train one or more machine learning models; using one or more test datasets to test one or more machine learning models; and tuning the parameters of multiple models to improve the accuracy of multiple models.

[0088] Example 18 is a method of Example 17, wherein applying one or more machine learning models to one or more datasets to train one or more machine learning models includes performing feature extraction on one or more datasets of sensor data.

[0089] Example 19 is a method of any of Examples 11 through 18, receiving one or more predictions from one or more machine learning models, including converting the predictions from one or more machine learning models into a format that can be displayed on a monitor.

[0090] Example 20 is a method of any of Examples 11 through 19, and also includes generating a graph that indicates the results from one or more machine learning models.

[0091] All features disclosed in the specification, including the claims, abstract, and drawings, as well as all steps in any disclosed method or process, may be combined in any combination, except that at least some of such features and / or steps are mutually exclusive combinations. Unless otherwise expressly stated, each feature disclosed in the specification (including the claims, abstract, and drawings) may be replaced by an alternative feature for the same, equivalent, or similar purpose.

[0092] Furthermore, specific features are mentioned in this written description. It should be understood that the disclosure in this specification includes all possible combinations of these specific features. Where a specific feature is disclosed in the context of a particular aspect or example, that feature may also be used, to the extent possible, in the context of other aspects and examples.

[0093] Furthermore, when a method having two or more defined steps or operations is mentioned in this application, the defined steps or operations may be performed in any order or simultaneously, unless the context precludes such possibilities.

[0094] Although specific examples of the invention have been shown and described for illustrative purposes, it will be understood that various modifications may be made without departing from the spirit and scope of the invention. Therefore, the invention should not be limited to anything other than the appended claims.

Claims

1. A testing and measuring instrument, comprising: One or more ports are used to connect to the hardware under test (HUT); A set of sensors connected to the HUT and the instrument; A display for showing one or more signal representations from at least one of the HUTs and from one or more sensors of the sensor group; and One or more processors are configured to execute code to cause the one or more processors to: Data is acquired from the sensor array; One or more datasets are formed based on the data acquired from the sensor group; Apply one or more machine learning models to the one or more datasets; and Receive predictive analytics about the HUT from the one or more machine learning models.

2. The testing and measuring instrument as described in claim 1, wherein, The sensor array connected to the instrument is connected to the instrument via a sensor data collector comprising either a microcontroller or a data logger.

3. The testing and measuring instrument according to claim 1, wherein, The code that enables the one or more processors to acquire data includes code that enables the one or more processors to acquire data from one of the sensors or from a remote memory.

4. The testing and measuring instrument as described in claim 1, wherein, The code that causes the one or more processors to form one or more datasets includes causing the one or more processors to normalize the data and form one or more training datasets, one or more test datasets, and one or more prediction datasets.

5. The testing and measuring instrument as described in claim 1, wherein, The code that causes the one or more processors to apply one or more machine learning models to the one or more datasets includes code that causes the one or more processors to perform the following operations: The system receives input through a user interface to identify the selected model to be applied to the one or more datasets; and Apply the selected model to one or more datasets.

6. The testing and measuring instrument as described in claim 5, wherein, Identifying the input to the selected model includes automatic selection, and the code that enables the one or more processors to apply machine learning includes code that enables the one or more processors to apply a machine learning model with the lowest error rate.

7. The testing and measuring instrument as described in claim 1, wherein, The one or more processors are also configured to execute code that enables the one or more processors to dynamically train one or more of the machine learning models at runtime using data streaming from the sensors.

8. The testing and measuring instrument as described in claim 1, wherein, The code that causes the one or more processors to apply machine learning to the dataset includes code that causes the one or more processors to perform the following operations: The one or more machine learning models are applied to one or more training datasets to train the one or more machine learning models; Use one or more test datasets to test the one or more machine learning models; and Adjust the parameters of the multiple models to improve their accuracy.

9. The testing and measuring instrument as described in claim 8, wherein, The code that causes the one or more processors to apply the one or more machine learning models to one or more datasets to train the one or more machine learning models includes code that causes the one or more processors to perform feature extraction on the one or more datasets of sensor data.

10. The test and measurement system of claim 1, wherein, The one or more processors are also configured to execute code to cause the one or more processors to generate graphs indicating results from the one or more machine learning models.

11. A method comprising: Receive data from a set of sensors connected to the hardware under test (HUT); One or more datasets are formed based on the data acquired from the sensor group; Apply one or more machine learning models to the one or more datasets; and Receive one or more predictions about the probability of failure of the HUT from the one or more machine learning models.

12. The method according to claim 11, wherein, Receiving data includes receiving the data from one of the sensors or from a remote memory.

13. The method of claim 11, wherein, Forming one or more datasets includes normalizing the data and forming one or more training datasets, one or more test datasets, and one or more prediction datasets.

14. The method of claim 11, wherein applying one or more machine learning models to the one or more datasets comprises: The user interface receives input identifying the selected model to be applied to the one or more datasets; and Apply the selected model to one or more datasets.

15. The method of claim 11, wherein, Identifying the input to the selected model includes automatic selection, and applying the machine learning model includes applying the machine learning model with the lowest error rate.

16. The method of claim 11, further comprising dynamically training one or more of the machine learning models at runtime using data streamed from the sensor.

17. The method according to claim 11, wherein, Applying the machine learning to the dataset includes: The one or more machine learning models are applied to one or more training datasets to train the one or more machine learning models; The one or more machine learning models are tested using one or more test datasets; and Adjust the parameters of the multiple models to improve their accuracy.

18. The method of claim 17, wherein, Applying the one or more machine learning models to one or more datasets to train the one or more machine learning models includes performing feature extraction on the one or more datasets of sensor data.

19. The method of claim 11, wherein receiving one or more predictions from the one or more machine learning models comprises converting the predictions from the one or more machine learning models into a format that can be displayed on the display.

20. The method of claim 11, further comprising generating a graph indicating results from the one or more machine learning models.

Images