Predicting performance degradation using nonlinear characteristics

Abstract

A technique for predicting gradual performance degradation with nonlinear characteristics is described. This technique includes a method that involves inputting new data samples into a fault prediction model, wherein the model is trained using a labeled historical dataset, where corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples, wherein the corresponding training samples are clustered into multiple clusters, and wherein each cluster is associated with a normal score and an anomaly score. The method further includes outputting a classification associated with the new data sample based on a comparison of a first anomaly score of a first cluster comprising the new data sample with the average anomaly score of clusters having normal scores greater than the anomaly score.

Description

Technical Field

[0001] This disclosure relates to predictive modeling, and more specifically, to the prediction of performance degradation using nonlinear characteristics. Background Technology

[0002] Nonlinear performance degradation can include phenomena associated with the gradual deterioration of one or more mechanical systems or components, where such deterioration may eventually lead to failure or permanent suboptimal performance of the one or more mechanical systems or components. Such gradual deterioration associated with nonlinear performance degradation can involve any type of wear (e.g., weakening through the gradual removal or deformation of a component based on its interaction with another material), fatigue (e.g., weakening due to periodic loading), creep (e.g., deformation due to sustained mechanical stress), and / or other nonlinear phenomena. Nonlinear performance degradation can be caused by mechanical, chemical, thermal, or other stresses. For example, wear phenomena can include abrasive wear, corrosive wear, erosive wear, and other types of wear.

[0003] However, predicting nonlinear performance degradation, such as wear-induced deterioration, presents several challenges. First, wear is a gradual failure that develops over extended periods. Therefore, the relationship between normal and wear states is nonlinear, rendering linear models (e.g., Natural Bayes, Support Vector Machines (SVMs), etc.) inapplicable. Furthermore, the progression of wear failure among similar assets is highly variable (e.g., some assets fail in 50 days while others fail in 6 months), depending on factors such as usage characteristics. As an example, wear-related performance degradation in a progressing cavity pump (PCP) can depend on a variety of factors, such as the subsurface geological formation type, sand grain size, and / or the operating profile of any particular PCP. Therefore, the level of performance degradation as a function of time-to-failure can vary across different assets at the same point in time based on various operating factors.

[0004] Another challenge in predicting nonlinear performance degradation involves the imbalance between normal and faulty operating states of assets experiencing nonlinear performance degradation (e.g., wear-related performance degradation). In other words, the majority (e.g., greater than 90%) of the training dataset includes normal performance, while very little data indicates faulty performance. Making accurate predictions from imbalanced training data is extremely difficult. For example, leveraging highly parameterized nonlinear methods (e.g., artificial neural networks (ANNs)) is not a reasonable solution within the range where a large number of parameters in an ANN can be finely tuned and optimized, requiring abundant data (sufficient examples with both classes). However, in cases of highly imbalanced data, the number of examples in outlier states is insufficient to effectively train an ANN. To remedy the problem of imbalanced data, oversampling and / or undersampling methods can be employed; however, these methods ultimately skew the original distribution in the data, thus biasing the solution (e.g., reducing accuracy due to increased false positives).

[0005] Another challenge related to accurately predicting nonlinear performance degradation is that nonlinear performance degradation, such as wear-related degradation, is not necessarily a catastrophic failure. Returning to the PCP example, while a worn PCP will operate suboptimally due to wear on the rotor blades, it does not necessarily stop operating. This makes the "failure date" subjective for the operator's decision to replace a pump with a given level of performance degradation (as opposed to a catastrophic failure date). This, in turn, complicates data labeling (e.g., determining when to label data from a PCP as a failure when the PCP actually continues to operate below standard performance).

[0006] Another challenge associated with predicting nonlinear performance degradation in real-world applications involves correctly identifying fault characteristics. This challenge is twofold. First, the available data must be evaluated to identify fault characteristics. This may involve inferring information from data that is not necessarily directly related to nonlinear performance degradation. For example, few assets tend to be explicitly equipped to directly measure wear-related performance degradation. Returning to the PCP example, the interaction between speed, productivity, torque, and casing pressure may be the only available information from which fault characteristics can be inferred. Second, the exclusivity of the identified fault characteristics must be evaluated. In other words, the identified fault characteristics may be associated with two or more phenomena, increasing the likelihood of false positives. Based on the above, it can be seen that nonlinear performance degradation presents further challenges: (i) identifying fault characteristics from the available data on nonlinear performance degradation, and (ii) determining whether the identified fault characteristics exclusively (or primarily) represent nonlinear performance degradation.

[0007] The combination of these problems makes predictive modeling of nonlinear performance degradation difficult. Therefore, techniques for accurately predicting performance degradation caused by nonlinear phenomena are needed. Summary of the Invention

[0008] This invention relates to a computer-implemented method comprising inputting new data samples into a fault prediction model. The fault prediction model is trained using a labeled historical dataset. Corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples. The corresponding training samples are clustered into multiple clusters, each cluster being associated with a normal score and an anomaly score. The method further comprises outputting a classification associated with the new data sample based on a comparison of a first anomaly score of a first cluster comprising the new data sample with the average anomaly score of clusters having normal scores greater than the anomaly score.

[0009] Advantageously, the present invention can preferably accurately predict nonlinear performance degradation from imbalanced training datasets. Furthermore, normal scores and outlier scores can preferably advantageously quantify the degree of anomalousness associated with classification. Another advantage of the present invention is that it preferably does not rely on ANNs (which tend to over-parameterize imbalanced training data) or on oversampling or undersampling techniques for imbalanced training data (which tend to shift classification).

[0010] According to a preferred embodiment, this disclosure also includes a retrospective window that defines the amount of sequentially preceding data points to be included in each respective training sample. Advantageously, the size of the feature markers indicating normal or abnormal operation can be adjusted. For example, a shorter retrospective window may be more sensitive to individual data points, while a longer retrospective window may be less sensitive to individual data points.

[0011] According to a preferred embodiment, the method further includes a prediction time domain defining a predefined amount of future time, wherein a corresponding label for a corresponding data point within the predefined future time domain is associated with a corresponding training sample. Advantageously, the prediction time domain can link various feature labels defined by a review window with corresponding future outcomes. For example, a shorter prediction time domain can provide a shorter warning for a given prediction (e.g., one day before a wear-related failure), while a longer prediction time domain can provide a longer warning for a given prediction (e.g., one month before a wear-related failure).

[0012] According to a preferred embodiment, the method further includes using K-means clustering to cluster the corresponding training samples. Advantageously, K-means clustering is an efficient and scalable clustering technique.

[0013] According to one aspect, a system is provided, comprising: one or more processors; and one or more computer-readable storage media storing program instructions configured, when executed by the one or more processors, to cause the one or more processors to perform a method comprising: inputting a new data sample into a fault prediction model, wherein the fault prediction model is trained using a labeled historical dataset, wherein corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples, wherein the corresponding training samples are clustered in a plurality of clusters, and wherein each of the plurality of clusters is associated with a normal score and an anomaly score; and outputting a classification associated with the new data sample based on comparing a first anomaly score of a first cluster containing the new data sample in the plurality of clusters with an average anomaly score of clusters in the plurality of clusters having a normal score greater than the anomaly score.

[0014] According to one aspect, a computer program product is provided, comprising one or more computer-readable storage media and program instructions commonly stored on the one or more computer-readable storage media, the program instructions including instructions configured to cause one or more processors to perform a method comprising: inputting a new data sample into a fault prediction model, wherein the fault prediction model is trained using a labeled historical dataset, wherein corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples, wherein the corresponding training samples are clustered in a plurality of clusters, and wherein each of the plurality of clusters is associated with a normal score and an anomaly score; and outputting a classification associated with the new data sample based on comparing a first anomaly score of a first cluster including the new data sample in the plurality of clusters with a normal score greater than the anomaly score in the plurality of clusters.

[0015] Another aspect of this disclosure relates to a computer-implemented method for predicting wear-related degradation of a screw pump (PCP), the method comprising inputting new data samples of PCP into a model configured to predict wear-related degradation of PCP. The model is trained using a labeled historical PCP dataset. Relevant data points are associated with a review window and a prediction time domain to create corresponding training samples. The corresponding training samples are clustered into multiple clusters, each cluster being associated with a normal score and an anomaly score. The method further includes outputting a classification associated with the new data sample based on a comparison of a first anomaly score of a first cluster comprising the new data sample with the average anomaly score of clusters having normal scores greater than the anomaly score, wherein the classification indicates wear-related degradation of the PCP.

[0016] Advantageously, the foregoing aspects of this disclosure can preferably accurately predict the nonlinear performance degradation of wear-related deterioration in PCP from an imbalanced training dataset of PCP-related data. Furthermore, normal scores and outlier scores can preferably advantageously quantify the degree of anomaly associated with the classification.

[0017] Another aspect of this disclosure relates to a computer-implemented method for predicting wear-related degradation of a screw pump (PCP), the method comprising generating labeled historical data by performing binary tagging of historical data associated with one or more PCPs. The method further comprises generating multiple training data samples by applying a review window and a prediction time domain to corresponding data points of the labeled historical data. The method further comprises clustering the multiple training data samples into multiple clusters. The method further comprises calculating a cluster score for each corresponding cluster within the multiple clusters. The method further comprises assigning new data samples of PCPs to a first cluster within the multiple clusters. The method further comprises assigning a classification to the new data samples based on the cluster score associated with the first cluster, wherein the classification indicates the likelihood of future wear-related degradation of the PCP.

[0018] Advantageously, this disclosure can preferably accurately predict the nonlinear performance degradation of wear-related deterioration in PCP from an imbalanced training dataset of PCP-related data. Furthermore, cluster scores can preferably advantageously quantify the degree of anomaly associated with classification.

[0019] According to a preferred embodiment, the method further includes marking historical data as faulty within a predetermined time period prior to a known pump replacement date. Advantageously, this embodiment of the present disclosure provides a clear boundary between "normal" and "erroneous" data, whereas such a deterministic boundary may not necessarily exist elsewhere due to the nonlinear and asymptotic nature of wear-related degradation in performance.

[0020] According to a preferred embodiment, the method further includes historical data such as pump speed data, pump torque data, casing pressure data, productivity data, and maintenance records related to the PCP. Advantageously, this is data available for the PCP. In other words, by using this data to predict wear-related degradation in PCP performance, no additional data instruments are required.

[0021] According to a preferred embodiment, the method further includes calculating a cluster score for a corresponding cluster, which further includes calculating a normal score for a first cluster, wherein the normal score is a first proportion of training data samples associated with normal states in the first cluster divided by a second proportion of training data samples associated with normal states in a plurality of training data samples. Calculating the cluster score also includes calculating an anomaly score for the first cluster, wherein the anomaly score is a third proportion of training data samples associated with deteriorated states in the first cluster divided by a fourth proportion of training data samples associated with deteriorated states in a plurality of training data samples. Advantageously, calculating the normal score and the anomaly score quantifies the relative degree of anomalousness in the predicted classification, thereby adding accuracy and context to the classification.

[0022] According to a preferred embodiment, the method further includes generating a fault signal for new data samples, wherein the fault signal comprises the average anomaly score of the new samples over a predetermined number of previous data points. Advantageously, the fault signal represents a post-processing smoothing of the anomaly scores for a given data sample, thereby reducing false positives and / or noise in consecutive anomaly scores of the data stream.

[0023] Additional aspects of this disclosure relate to systems and computer program products configured to perform the methods described above. This invention is not intended to illustrate every aspect, implementation, and / or embodiment of every embodiment of this disclosure. Attached Figure Description

[0024] Preferred embodiments of the invention will now be described by way of example only and with reference to the following figures:

[0025] Figure 1 A block diagram of an example computing environment according to some embodiments of the present disclosure is shown.

[0026] Figure 2 A block diagram of an example fault prediction model according to some embodiments of the present disclosure is shown.

[0027] Figure 3 A flowchart is shown as an example method for predicting performance degradation with nonlinear characteristics according to some embodiments of the present disclosure.

[0028] Figure 4 A flowchart is shown of an example method for training a fault prediction model according to some embodiments of the present disclosure.

[0029] Figure 5A A flowchart is shown as an example method for generating results based on the output from a fault prediction model, according to some embodiments of the present disclosure.

[0030] Figure 5B A flowchart of an example method for calculating fault signals according to some embodiments of the present disclosure is shown.

[0031] Figure 6A Experimental results are shown in the graph of cumulative anomaly score as a function of the number of days since the fault, according to some embodiments of the present disclosure.

[0032] Figure 6B Experimental results are shown in the graphs of the intensity of the failure label as a function of the number of days since the failure, according to some embodiments of the present disclosure.

[0033] Figure 7 Experimental results of confusion matrices at various time points prior to failure are shown according to some embodiments of this disclosure.

[0034] Figure 8 A block diagram of an example computer according to some embodiments of the present disclosure is shown.

[0035] Figure 9 A cloud computing environment according to some embodiments of the present disclosure is described.

[0036] Figure 10 An abstract model layer according to some embodiments of this disclosure is described.

[0037] This disclosure can be modified in various ways and alternatives, the details of which are shown by way of example in the accompanying drawings and will be described in detail. However, it is to be understood that the purpose is not to limit this disclosure to the specific embodiments described. Rather, it is intended to cover all modifications, equivalents, and alternatives that fall within the spirit and scope of this disclosure. Detailed Implementation

[0038] Embodiments of this disclosure relate to predictive modeling, and more specifically, to predicting performance degradation using nonlinear characteristics. The embodiments of this disclosure will be better understood from the foregoing context, and are not limited to these applications.

[0039] Embodiments of this disclosure relate to techniques for detecting asset performance degradation during the early stages of nonlinear failure mechanisms (e.g., wear, fatigue, creep, etc.) to predict impending asset failure and recommend preventative maintenance before such failure. Embodiments of this disclosure utilize semi-supervised machine learning methods that ingest historical data about the asset as input and generate anomaly scores, classifications, and / or failure signals as outputs indicating the likelihood of future asset failure or performance degradation.

[0040] In overcoming the challenges previously discussed in predictive modeling of nonlinear phenomena, embodiments of this disclosure achieve features such as, but not limited to, the following: (i) a nonlinear decision boundary that distinguishes normal data from erroneous data; (ii) a computationally straightforward implementation (e.g., although nonlinear, it is not over-parameterized as in the case of ANNs); (iii) no class imbalance (therefore, the original distribution in the data is not skewed and the subsequent solution is not shifted); and / or (iv) it can be readily applied to global solutions for any similar family of assets.

[0041] Now refer to the attached diagram, Figure 1 An example computing environment 100 according to some embodiments of the present disclosure is illustrated. The computing environment 100 includes a fault prediction system 102 communicatively coupled to a sensor data logging system 104 and an endpoint application 106 via a network 108. The fault prediction system 102 may be configured to receive data from the sensor data logging system 104 and make predictions related to future performance degradation of the associated asset 110 due to nonlinear phenomena (e.g., predictions of wear-related performance degradation of PCP over future time intervals). The fault prediction system 102 may be further configured to interact with the endpoint application 106. For example, the fault prediction system 102 may receive a request from the endpoint application 106 to detect any indication of a fault from the sensor data logging system 104. As another example, the fault prediction system 102 may push updates about potential fault characteristics identified from the sensor data logging system 104 to the endpoint application 106.

[0042] Sensor data logging system 104 may include one or more data acquisition systems configured to acquire data directly or indirectly related to the functionality of asset 110. For example, in the case where asset 110 is a PCP in an oil well production system, sensor data logging system 104 may collect data such as, but not limited to, speed, torque, casing pressure, and / or productivity. As another example, in the case where asset 110 is a mechanical component of a vehicle (e.g., fuel pump, wheel bearings, cylinder head gaskets, etc.), sensor data logging system 104 may collect data such as, but not limited to, mileage, speed, engine error codes, etc. As can be seen from these two non-limiting examples, sensor data logging system 104 does not necessarily need to collect data directly associated with asset 110. Instead, in some embodiments, sensor data logging system 104 collects data associated with other components related to asset 110, but the collected data may still be used to provide indirect inferences about the functionality of asset 110. This may be beneficial where using specific components of an equipment system to collect data to predict nonlinear performance degradation may be economically infeasible and / or technically impractical. Therefore, in many real-world applications, predicting nonlinear performance degradation presents the following challenges: (i) identifying fault features from available data on nonlinear performance degradation, and (ii) determining whether the identified fault features exclusively (or primarily) represent nonlinear performance degradation.

[0043] Endpoint application 106 may be an application running on a user workstation such as, for example, a desktop computer, laptop computer, tablet computer, smartphone, or other endpoint device. Endpoint application 106 may provide an interface for a user to interact with fault prediction system 102. For example, a user may request a predicted fault analysis of asset 110 based on data from sensor data logging system 104. As another example, fault prediction system 102 may push updates, notifications, or warnings to endpoint application 106 based on fault characteristics associated with asset 110 and detected from data from sensor data logging system 104. Furthermore, in some embodiments, endpoint application 106 provides a mechanism for a user to configure the trained fault prediction system 102 to receive streaming data, where the streaming data can be used on assets similar to asset 110 used to train fault prediction system 102 (e.g., an oil well operator may stream their own PCP data to fault prediction system 102 previously trained on similar PCP data from one or more other PCPs).

[0044] In some embodiments, the fault prediction system 102 may be provided virtually within a cloud computing architecture. In some embodiments, the fault prediction system 102 may reside in a computer, such as, for example, a mainframe, compute node, desktop, laptop computer, tablet computer, or another system including one or more processors and one or more computer-readable storage media.

[0045] The fault prediction system 102 may include a data warehouse 112, a model container 114, and a computing engine 116. The data warehouse 112 may include aggregated sensor data 118, which may be data collected from the sensor data recording system 104 and may include one or more data samples.

[0046] Model container 114 may include a data preparation / engineering utility 126, which can be executed on aggregated sensor data 118 to at least partially generate formatted sensor data 120. For example, the data preparation / engineering utility 126 may be configured to remove outliers, correct data formatting issues, resolve null values, etc., while converting aggregated sensor data 118 into formatted sensor data 120. In some embodiments, the formatted sensor data 120 may, for example, include a look-back window applied to the aggregated sensor data 118.

[0047] Formatted sensor data 120 can be input into fault prediction model 124. In some embodiments, the formatted sensor data 120 and fault prediction model 124 are loaded into deployment resources 130 of computing engine 116. After execution of fault prediction model 124 using formatted sensor data 120 as input, computing engine 116 can generate results 122 and store results 122 in data warehouse 112. Results 122 may include, for example, anomaly scores, classifications, and / or fault signals. Results 122 may indicate the likelihood of future performance degradation due to nonlinear performance degradation (e.g., the likelihood of wear-related performance degradation in PCP).

[0048] The computing engine 116 may further include a prediction service 128, which may be configured to receive requests from or push notifications to the endpoint application 106. The prediction service 128 may coordinate the functionality of the fault prediction system 102. For example, in some embodiments, the prediction service 128 may enable the use of deployment resources 130 to perform data preparation / engineering utilities 126 on aggregated sensor data 118 to generate formatted sensor data 120. Continuing with the above example, the prediction service 128 may be further configured to deploy a fault prediction model 124 on deployment resources 130 and use the formatted sensor data 120 as input to generate a result 122. The prediction service 128 may also be configured to send the result 122 to the endpoint application 106.

[0049] Now go to Figure 2 A block diagram of a fault prediction model 124 according to some embodiments of the present disclosure is shown. The fault prediction model 124 may include, for example, a training mode 200 and a deployment mode 228. In training mode 200, the fault prediction model 124 is trained to accurately predict nonlinear performance degradation, such as faults or degradation due to wear. In deployment mode 228, the fault prediction model 124 is configured to receive input data, format the input data, and make predictions related to nonlinear phenomena associated with the data (e.g., predictions of future suboptimal performance due to wear in the PCP).

[0050] In training mode 200, fault prediction model 124 may receive historical data 202 from one or more sensor data recording systems 104 associated with one or more assets 110, wherein the historical data 202 is divided into a training subset 204-1 and a test subset 204-2. The training subset 204-1 may be configured to train and validate fault prediction model 124, and the test subset 204-2 may be used to test fault prediction model 124, wherein the test may be used to quantify the characteristics of fault prediction model 124, such as, for example, accuracy, precision, recall, etc.

[0051] Training mode 200 may also include labeled historical data 206, which can be derived from aligning historical fault records (e.g., maintenance records indicating PCP replacement) with sensor readings (e.g., speed, torque, productivity, casing pressure, etc.) over time. Labeled historical data 206 includes data 208-1 indicating normal operation of asset 110 and data 208-2 indicating deterioration of asset 110's suboptimal, degraded, or faulty performance (e.g., PCP wear condition). The decision boundary between data 208-1 indicating normal operation and data 208-2 indicating deterioration can be determined subjectively by a subject matter expert (SME), objectively by statistical measurements (e.g., beyond one standard deviation of the mean during normal operation), inferred from machine learning, or using other strategies or techniques. As previously discussed, data 208-2 indicating deterioration is not necessarily associated with a non-operating asset 110. Conversely, data 208-2 indicating a deterioration state indicates suboptimal performance of asset 110 even if it remains operational. For example, in some embodiments, if asset 110 experiences a significant failure or is otherwise replaced on day x, data from a predetermined period prior to day x can be automatically labeled as data 208-2 indicating a deterioration state.

[0052] Training mode 200 may also include windowed historical data 210. Windowed historical data 210 may include a look-back window 212 and a prediction horizon 214. The look-back window 212 may refer to multiple data points preceding the current data point (e.g., D...). x This is included in each data sample. The prediction time domain 214 can refer to the labels of data points, i.e., multiple future data points starting from the current data point (e.g., D). yThe review window 212 and the prediction time domain 214 can be used to transform the corresponding data points in the training subset 204-1 into corresponding data samples 216-1 associated with the normal state and corresponding data samples 216-2 associated with the deteriorated state (collectively referred to as data samples 216). In other words, the failure prediction model 124 is trained to use the performance history of asset 110 (equal to the review window 212) to draw conclusions about the possible future states of asset 110 regarding the level of performance degradation. The corresponding data samples 216 may include vectors, matrices, or tensors of data points corresponding to the current data point and several previous data points (based on the review window 212) for each data stream, and the corresponding data samples 216 can also be associated with labels of future data points through the prediction time domain 214 (where the labels of future data points can be added to predetermined positions in the vector, matrix, or tensor, or otherwise associated with the vector, matrix, or tensor). In other words, the data sample 216-1 associated with the normal state can be identified from the data point at prediction time domain 214; this data point is data 208-1 indicating the normal state. Similarly, the data sample 216-2 associated with the deterioration state can be those data samples that have data points marked as data 208-2 indicating the deterioration state at prediction time domain 214.

[0053] For example, the review window 212 could be ten days, and the prediction time domain 214 could be twenty days. In this scenario, for a data point at a first time from a single data stream in the training subset 204-1, the data point can be transformed into a vector including the first ten data points (e.g., review window 212), and this vector can be associated with labels for data points in the next twenty days (e.g., prediction time domain 214, where the label is data 208-1 indicating a normal state or data 208-2 indicating a deteriorated state). In this example, if the label for the data points in the next twenty days is data 208-2 indicating a deteriorated state, then the data point vector of the current data point and the first ten data points can be considered a predicted fault feature for training purposes. Conversely, if the label for the data points in the next twenty days is data 208-1 indicating a normal state, then the data point vector of the current data point and the first ten data points can be assumed to be a predicted normal feature for training purposes.

[0054] As those skilled in the art will understand, the review window 212 and the prediction time domain 214 can be various numbers based on various scales. For example, in some embodiments, the review window 212 and the prediction time domain 214 can be measured in seconds, minutes, days, weeks, months, etc. In various embodiments, the review window 212 is less than, greater than, or equal to the prediction time domain 214. In some embodiments, the size of the review window 212 is determined as a trade-off between administrative utility and computational overhead. For example, a relatively larger review window 212 provides increased information to accurately detect fault characteristics while requiring additional computational power. Conversely, a relatively smaller review window 212 provides reduced information to accurately detect fault characteristics while requiring less computational power. In some embodiments, the size of the prediction time domain 214 is determined based on fault characteristics. For example, if asset 110 is associated with a twenty-day window from the time when the fault indication begins to be indicated in the data, then the prediction time domain 214 must be twenty days or less (in other words, a prediction time domain 214 greater than twenty days would result in increased false positives). Furthermore, in some embodiments, the review window 212 and the prediction time domain 214 do not need to be measured in time-based increments at all, but can simply be referenced by multiple previous or subsequent data points, where the interval between data points can be based on non-temporal characteristics. Additionally, although the data sample 216 discussed in the above example is in vector format, in other embodiments, matrices or tensors can be used to represent multidimensional data or multimodal data streams. As an example, for asset 110 associated with three data streams, data sample 216 may include input-output data pairs, where the input portion includes a three-dimensional tensor consisting of corresponding review window 212 samples from each of the three data streams, and where the output portion includes a binary indicator of future performance at the prediction time domain 214 (e.g., 0 for normal, 1 for fault).

[0055] Training mode 200 may also include clustered data 218. Clustering 218 may use any testable clustering technique now known or developed in the future to cluster the data samples 216. In other words, the clustered data 218 may include multiple clusters 220, where each cluster includes at least one data sample 216. It is worth noting that the number of clusters 220 is configurable in order to accurately fit (without overfitting) the data samples 216.

[0056] In some embodiments, cluster 220 is determined by using K-means clustering. Advantageously, K-means clustering is a computationally efficient clustering technique that can scale to large datasets. More generally, according to some embodiments of the present disclosure, the type of clustering technique used, the number of clusters used, and the parameters of the clusters used (e.g., shape, size, etc.) are all adjustable parameters that can be adjusted as needed to improve the performance of the fault prediction model 124. For example, it may be beneficial to have a sufficient number of clusters to capture various normal operation profiles and various degraded operation profiles.

[0057] The clustered data 218 may also include a cluster score 222 assigned to each of the clusters in the cluster 220. The cluster score 222 may include a normality score 224 and an anomaly score 226. In some embodiments, the normality score 224 can be calculated according to Equation 1:

[0058] Equation 1: For 1 < i < the number of clusters

[0059] Similarly, in some embodiments, the anomaly score 226 can be calculated according to Equation 2:

[0060] Equation 2: For 1 < i < the number of clusters

[0061] Regarding Equation 1, may refer to the count of data samples 216-1 associated with the normal state within a given cluster C of the cluster 220, while N i may refer to the total count of data samples 216-1 associated with the normal state in the windowed historical data 210. Similarly, regarding Equation 2, the term n may refer to the count of data samples 216-2 associated with the degraded state within a given cluster C of the cluster 220, while N may refer to the total count of data samples 216 associated with the degraded state in the windowed historical data 210. For both Equation 1 and Equation 2, the term i may refer to the total count of data samples 216 within a given cluster C of the cluster 220, while N may refer to the total count of data samples 216 in the windowed historical data 206.

[0062] ​​​​​​In other words, the normality score 224 can be the proportion of data samples 216-1 associated with the normal state in a given cluster divided by the proportion of data samples 216-1 associated with the normal state in the entire windowed historical data 210. Similarly, the anomaly score 226 can be the proportion of data samples 216-2 associated with the deteriorated state in a given cluster divided by the proportion of data samples 216-2 associated with the deteriorated state in the entire windowed historical data 210.

[0063] After creating cluster 220 and generating cluster scores 222, the fault prediction model 124 can be considered trained. In some embodiments, after training the fault prediction model 124 using training subset 204-1, the fault prediction model 124 can be tested using test subset 204-2 and deployment mode 228. Although deployment mode 228 will be discussed with respect to training subset 204-2, the discussion of deployment mode 228 is equally applicable to receiving streaming real-time data in order to predict future nonlinear phenomena associated with new data (e.g., wear-related performance degradation in PCP).

[0064] In deployment mode 228, fault prediction model 124 can format test subset 204-2 into windowed data samples 230. Windowed data samples 230 may resemble data sample 216, but without any indication of normal or deteriorated states (in this case, this is information predicted by fault prediction model 124 and therefore hidden from fault prediction model 124 when testing its performance). Therefore, the corresponding samples in windowed data samples 230 may include data points from test subset 204-2 and a previous number of data points based on review window 212. As previously mentioned, this set of data can be stored in vector, matrix, or tensor formats depending on the complexity, dimension, and form of the data in test subset 204-2. In some embodiments, windowed data samples 230 correspond to formatted sensor data 120.

[0065] Then, the corresponding sample in the windowed data sample 230 can be associated with the corresponding cluster 220. Classification 232 can then be associated with each windowed data sample 230 based on the associated clusters within cluster 220. For example, a windowed data sample 230 placed within a corresponding cluster 220 having a normal score 224 greater than the anomaly score 226 can be considered normal (e.g., a "0" score). Conversely, if a corresponding windowed data sample 230 is placed within a corresponding cluster 220 having an anomaly score 226 greater than the normal score 224, the corresponding windowed data sample 240 can be considered anomalous or predicted to fail in the future (e.g., a "1" score). In some embodiments, classification 232 includes comparing the anomaly score 226 of the cluster 220 that collected the windowed data sample 230 with the average anomaly score of all clusters 220 having a normal score 224 greater than the anomaly score 226. In this way, the anomaly score 226 of cluster 220 of the windowed data sample 230 can be compared with the baseline level of anomalies seen in other clusters 220 that are considered normal.

[0066] The fault prediction model 124 may also include a fault signal 234. The fault signal 234 can be configured to smooth out anomaly scores 226 or classifications 232 by acquiring the average of the anomaly scores or a previously predetermined number of data points. For example, the fault signal 234 can be calculated according to Equation 3:

[0067] Equation 3:

[0068] In Equation 3, the term x may refer to a predetermined number of sequentially preceding data points on which the average outlier score is determined for a given window of data samples 230. In some embodiments, x may be an adjustable parameter considered in the design of the fault prediction model 124. For example, a relatively larger x may reduce the sensitivity of the fault prediction model 124 to any particular outlier score indicating a fault (and thus reduce false positives), while a relatively smaller x may increase the sensitivity of the fault prediction model 124 to each outlier score indicating a fault (and thus reduce false negatives). Furthermore, in Equation 3, the term AS... n This can refer to the anomaly score 226 of the cluster 220 that includes data point n, although classification 232 may also be used in other embodiments. See reference... Figure 5B Fault signal 234 will be discussed in more detail.

[0069] Now for reference Figure 3The diagram illustrates a flowchart of an example method 300 for utilizing a fault prediction model 124 according to some embodiments of the present disclosure. Method 300 may be implemented by, for example, the fault prediction model 124, a fault prediction system 102, a computer, a computing node, a processor, or another combination of hardware and / or software.

[0070] Operation 302 includes training the fault prediction model 124. Training the fault prediction model 124 may involve aspects previously discussed with respect to training mode 200 of the fault prediction model 124. Reference will be made below. Figure 4 Let's discuss Operation 302 in more detail.

[0071] Operation 304 includes formatting the aggregated sensor data 118 into formatted sensor data 120. In some embodiments, the formatted sensor data 120 corresponds to windowed data samples 230. Operation 304 may include applying a review window 212 to the corresponding data points in the aggregated sensor data 118 to generate the formatted sensor data 120. In some embodiments, operation 304 also includes other data cleaning and / or data formatting operations, such as removing outliers, parsing null values, etc.

[0072] Operation 306 includes inputting formatted sensor data 120 into fault prediction model 124. Operation 308 includes generating result 122 based on the output from fault prediction model 124. In some embodiments, result 122 includes classification 232 (e.g., normal or indicating future faults) and / or fault signal 234 (e.g., quantification related to the probability of future faults).

[0073] Operation 310 includes performing mitigation actions based on the outcome. For example, mitigation actions may involve sending outcome 122 to endpoint application 106 in the form of a notification, warning, report, or other transmission. In some embodiments, operation 310 includes triggering a scheduling event related to the maintenance of asset 110, such as replacing, rebuilding, or otherwise maintaining asset 110. In some embodiments, the scheduling event is based on outcome 122. For example, a fault signal 234 exceeding a threshold may trigger a maintenance event to be scheduled within a time window (e.g., within the prediction time domain 214). In some embodiments, mitigation actions may be related to logistical actions, such as ordering any necessary replacement parts and sending any ordered replacement parts to the location associated with asset 110.

[0074] Now for reference Figure 4 The diagram illustrates a flowchart of an example method 400 for training a fault prediction model 124 according to some embodiments of the present disclosure. In some embodiments, method 400 is... Figure 3The sub-method of operation 302. In some embodiments, method 400 may be implemented, for example, by fault prediction model 124, fault prediction system 102, computer, computing node, processor, or another combination of hardware and / or software.

[0075] Operation 402 includes aggregating historical data 202 into a training subset 204-1 and a test subset 204-2. Operation 404 includes labeling historical data 202 to generate labeled historical data 206, including data 208-1 indicating a normal state and data 208-2 indicating a deteriorated state. Operation 406 includes generating data samples 216-1 associated with the normal state and data samples 216-2 associated with the deteriorated state by applying a review window 212 and a prediction time domain 214 to the corresponding data points in the historical data 202.

[0076] Operation 408 includes clustering the data sample 216 into multiple clusters 220. In some embodiments, operation 408 utilizes K-means clustering. Operation 410 includes calculating a cluster score 222 associated with each cluster in the clusters 220. The cluster score 222 may include a corresponding normal score 224 and a corresponding outlier score 226 for each respective cluster in the clusters 220.

[0077] Operation 412 includes adjusting the fault prediction model 124. Adjusting the fault prediction model 124 may include, for example, (i) modifying the tagged historical data 206 by changing the definitions of data 208-1 indicating a normal state and data 208-2 indicating a deteriorated state; (ii) changing the size of the review window 212; (iii) changing the size of the prediction time domain 214; (iv) changing the parameters associated with cluster 220 (e.g., the number of clusters, the shape of the clusters, the size of the clusters, etc.); and / or other model adjustment techniques and / or strategies. In some embodiments, the fault prediction model 124 is adjusted based on the results of the test subset 204-2.

[0078] Operation 414 includes outputting the trained fault prediction model 124. In some embodiments, outputting the trained fault prediction model 124 includes storing the trained fault prediction model 124 in a computer-readable storage medium, such as, for example, a virtually provided model container 114.

[0079] Now for reference Figure 5A The diagram illustrates a flowchart of an example method 500 for generating results based on the output from a fault prediction model 124, according to some embodiments of the present disclosure. In some embodiments, method 500 is... Figure 3 The sub-method of operation 308. In some embodiments, method 500 may be implemented, for example, by fault prediction model 124, fault prediction system 102, computer, computing node, processor, or another combination of hardware and / or software.

[0080] Operation 502 includes associating the binary classification result 232 with a corresponding data sample (e.g., windowed data sample 230) previously input into the fault prediction model 124. In some embodiments, classification 232 is "normal" (e.g., 0) or "abnormal," "faulty," or another abnormal indicator (e.g., 1). Classification 232 may be based on a normality score 224 and anomaly score 226 of the cluster 220 that collects the corresponding windowed data sample 230. More specifically, if the normality score 224 is greater than the anomaly score 226, the corresponding windowed data sample 230 is considered normal. Conversely, if the anomaly score 226 is greater than the normality score 224, the corresponding windowed data sample 230 is considered anomalous or otherwise indicates a fault.

[0081] Operation 504 includes generating a fault signal 234 for the corresponding input data sample (e.g., windowed data sample 230). The fault signal 234 may be based on a normality score 224, an anomaly score 226, and / or classification 232 associated with the windowed data sample 230. In some embodiments, the fault signal 234 represents a more reliable indicator of truly anomalous data (e.g., it reduces false positives). References are made below. Figure 5B Fault signal 234 will be discussed in more detail.

[0082] Now for reference Figure 5B A flowchart of an example method 510 for generating a fault signal 234 according to some embodiments of the present disclosure is shown. In some embodiments, method 510 is Figure 5A The sub-method of operation 504. In some embodiments, method 510 may be implemented by, for example, a fault prediction model 124, a fault prediction system 102, a computer, a computing node, a processor, or another combination of hardware and / or software.

[0083] Operation 512 includes calculating the average anomaly score for the clusters 220 ranked as normal. Calculating the average anomaly score during normal operation may involve averaging the anomaly scores 226 for each cluster in clusters 220, where the normal score 224 is greater than the anomaly score 226. This is beneficial in establishing a baseline anomaly level used to reduce false positives.

[0084] Operation 514 involves assigning the input windowed data sample 230 to a cluster 220 associated with anomaly score 226 and normal score 224. If the anomaly score of the specified cluster is greater than the average anomaly score during normal operation (as determined in operation 512), the data sample is classified as being in a failure mode (classified as 1); otherwise, the data sample is classified as being in a normal mode (classified as 0).

[0085] Operation 516 includes calculating the fault signal 234 as the average of binary 0 or 1 classification results for the asset over a predetermined time period (e.g., 10 days) or a predetermined number of windowed data samples 230. In some embodiments, operation 516 may utilize Equation 3.

[0086] Refer again Figure 1 -5. One particular application of aspects of this disclosure relates to detecting performance degradation caused by wear on the rotor in a screw pump (PCP). Since the performance degradation caused by wear on the rotor in a PCP is a non-linear phenomenon, aspects of this disclosure are well-suited for accurately predicting rotor degradation caused by wear in a PCP.

[0087] Artificial lift systems utilizing PCPs enable various non-thermal oil and gas recovery methods, such as cold heavy oil production using sand (CHOPS). Compared to other artificial lift systems (e.g., electric submersible pumps (ESPs)), PCPs can lift viscous mixtures of oil and sand from underground reservoirs to the surface with improved lift costs, improved maintenance costs, improved application flexibility, and reduced environmental impact.

[0088] Although PCPs are suitable for handling heavy oil with high sand content, a problem caused by continuous sand intake is wear failure. Abrasive wear refers to the gradual degradation of pump performance as the hard chrome plating on the rotor wears away, and it is the most common type of failure in PCPs. This wear can be confined to the chrome-plated surface on the rotor or extend to the base metal. In either case, the original rotor profile is altered. This change in profile affects the performance of the PCP because it changes the fit between the rotor and stator. Severe wear from the hard chrome plating to the base metal can permanently damage the elastomers in the rotor, requiring pump replacement.

[0089] Pump failures in oil wells are costly in terms of lost production time. Therefore, the ability to predict pump wear failures mitigates these costs by providing proactive, scheduled maintenance for the PCP (Potentially Specific Pump) before failure occurs. Furthermore, awareness of improvements in early-stage wear-related pump performance degradation can help operators make appropriate adjustments to operational decisions to extend service life.

[0090] However, predicting failures in a PCP is challenging for reasons similar to predicting any performance degradation with nonlinear characteristics. For example, failure mechanisms are gradual, which raises problems in correctly labeling historical data as “normal” or “abnormal”, as the PCP may remain operating at suboptimal performance over extended periods while the rotor becomes increasingly worn. Another challenge involves the unbalanced set of historical data available for the PCP (e.g., significantly more normal data than anomalous data). Yet another challenge involves variable failure mechanisms that vary with the operating environment (e.g., including geological formations with higher sand content compared to those with lower sand content). Therefore, accurately predicting PCP failures is difficult. Nevertheless, when the aspects of this disclosure are implemented in the field of predictive monitoring of PCPs, PCP performance degradation due to wear-related mechanisms can be accurately predicted.

[0091] For example, returning again Figure 4 The discussion, because it involves training a failure prediction model 124 for asset 110 (e.g., PCP), may involve operation 402 aggregating sensor data, such as pump speed, pump torque, bushing pressure, and productivity. The aggregated sensor data may be divided into a training subset 204-1 (e.g., approximately 80% of the data) and a test subset 204-2 (e.g., approximately 20% of the data). Operation 404 may label historical data as either data 208-1 indicating a normal state (approximately 97% of the unbalanced data in training subset 204-1) or data 208-2 indicating a deteriorated state (approximately 3% of the unbalanced data in training subset 204-1). In some embodiments, data 208-2 indicating a deteriorated state may be any data between 25 days and 3 days prior to a known PCP replacement date (e.g., determined from maintenance records), while data prior to 25 days prior to the known PCP replacement date may be considered data 208-1 indicating a normal state.

[0092] Operation 406 generates data samples 216 by applying the review window 212 and the prediction time domain 214 to the corresponding data points. Operation 408 clusters the data samples 216 using K-means clustering, and operation 410 calculates the normality score 224 and the anomaly score 226 according to Equations 1 and 2, respectively. Operation 412 adjusts the fault prediction model 124, and operation 414 outputs the trained fault prediction model 124.

[0093] Applying the test subset 204-2 to the trained fault prediction model 124 for PCP applications demonstrates the utility of various aspects of this disclosure. Figure 6AThe experimental results are shown as a graph of the cumulative anomaly score 226 (y-axis) as a function of the number of days until the known pump replacement date (x-axis). It can be seen that the trend line increases sharply between 25 days before the failure and the known pump replacement date. Therefore, Figure 6A Anomaly score 226 is shown to be useful for successfully predicting PCP faults.

[0094] Figure 6B Experimental results are shown as a graph illustrating the intensity of fault category 232 as a function of days to failure (x-axis). It can be seen that the intensity of fault category 232 increases significantly between approximately 25 days prior to failure and the known pump replacement date. Therefore, Figure 6B It has been demonstrated that classification 232 can be successfully used to predict PCP faults.

[0095] It is worth noting that although no graph related to fault signal 234 is provided, it will be apparent to those skilled in the art that fault signal 234 is relative to... Figure 6A and 6B The results shown demonstrate similar predictive ability with reduced noise. This is because the fault signal 234 represents the average score, while... Figure 6A The cumulative score is shown. Figure 6B The intensity score is shown.

[0096] Figure 7 Experimental results related to the confusion matrix are shown for various time periods prior to the known pump replacement date, using test subset 204-2 for the PCP failure prediction model 124 as described above. Figure 7 As shown, 30 days before the failure, the ratio for the normal (actual label 702-1) - normal (predicted label 704-1) box was 0.76, the failure-normal box was 0.38, the normal-failure box was 0.24, and the failure-failure box was 0.62. 14 days before the failure, for 700-2, the ratio for the normal (actual label 702-2) - normal (predicted label 704-2) box was 0.76, the failure-normal box was 0.26, the normal-failure box was 0.24, and the failure-failure box was 0.74. 5 days before the failure, for 700-3, the ratio for the normal (actual label 702-3) - normal (predicted label 704-3) box was 0.76, the failure-normal box was 0.12, the normal-failure box was 0.24, and the failure-failure box was 0.88.

[0097] generally, Figure 7 Improved predictive performance is shown as a function of the approximation of the actual pump replacement date. Furthermore, Figure 7The results show recall rates of approximately 88% for 700-3 days before the failure, approximately 75% for 700-2 days before the failure, and approximately 62% for 700-1 days before the failure (e.g., true positives divided by the sum of true positives and false negatives). Meanwhile, aspects of this disclosure achieve precision of approximately 78% (700-3 days before the failure), 76% (700-2 days before the failure), and 72% (700-1 days before the failure) (e.g., true positives divided by the sum of true positives and false positives). Therefore, Figure 7 It has been demonstrated that various aspects of this disclosure realize a robust fault prediction model 124 for predicting PCP performance degradation due to rotor wear.

[0098] Figure 8 A block diagram of an example computer 800 according to some embodiments of the present disclosure is shown. In various embodiments, the computer 800 can perform... Figure 3 Any or all of the methods and / or implementations described in -5 Figure 1-2 And / or the functions discussed in any of 6-7. In some embodiments, computer 800 receives instructions related to the methods and functions described above by downloading processor-executable instructions from a remote data processing system via network 850. In other embodiments, computer 800 provides instructions for the methods and / or functions described above to a client machine, causing the client machine to perform the method or a portion thereof based on the instructions provided by computer 800. In some embodiments, computer 800 is incorporated into (or similar functions of computer 800 are virtually provided to) [the system / system]. Figure 1 Fault prediction system 102 Figure 1 Fault prediction model 124 or another aspect of this disclosure.

[0099] Computer 800 includes memory 825, storage device 830, interconnect 820 (e.g., bus), one or more CPUs 805 (also referred to herein as processors), I / O device interface 810, I / O device 812, and network interface 815.

[0100] Each CPU 805 retrieves and executes programming instructions stored in memory 825 or storage device 830. Interconnect 820 is used to move data, such as programming instructions, between CPU 805, I / O device interface 810, memory 830, network interface 815, and memory 825. Interconnect 820 may be implemented using one or more buses. In various embodiments, CPU 805 may be a single CPU, multiple CPUs, or a single CPU with multiple processing cores. In some embodiments, CPU 805 may be a digital signal processor (DSP). In some embodiments, CPU 805 includes one or more 3D integrated circuits (3DICs) (e.g., 3D wafer-level package (3DWLP), 3D intermediate-based integration, 3D stacked IC (3D-SIC), monolithic 3D IC, 3D heterogeneous integration, 3D system-in-package (3DSiP), and / or package-on-package (PoP) CPU configurations). Memory 825 is typically included to represent random access memory (e.g., static random access memory (SRAM), dynamic random access memory (DRAM), or flash memory). This typically includes memory 830 to represent non-volatile memory, such as hard disk drives, solid-state drives (SSDs), removable memory cards, optical storage devices, or flash memory devices. In an alternative embodiment, storage 830 may be replaced by a storage area network (SAN) device, the cloud, or other device connected to computer 800 via I / O device interface 810 or to network 850 via network interface 815.

[0101] In some embodiments, memory 825 stores instructions 860. However, in various embodiments, instructions 860 are partially stored in memory 825 and partially stored in storage device 830, or they are entirely stored in memory 825 or entirely stored in storage device 830, or they are accessed via network 850 through network interface 815.

[0102] Instruction 860 can be used to execute Figure 3 -5 any part or all and / or implementation of the methods. Figure 1-2 And / or computer-readable and computer-executable instructions for the functions discussed in 6-7. In some embodiments, instruction 860 may be referred to as a nonlinear performance degradation prediction protocol (or instruction, mechanism, etc.) or a fault prediction protocol (or instruction, mechanism, etc.). Although instruction 860 is shown in memory 825, instruction 860 may include program instructions that are co-stored across multiple computer-readable storage media and executable by one or more CPUs 805.

[0103] In various embodiments, I / O device 812 includes an interface capable of presenting information and receiving input. For example, I / O device 812 may present information to a user interacting with computer 800 and receive input from the user.

[0104] Computer 800 is connected to network 850 via network interface 815. Network 850 may include physical, wireless, cellular, or other networks.

[0105] It is understood that although this disclosure includes a detailed description of cloud computing, the implementation of the teachings set forth herein is not limited to a cloud computing environment. Rather, embodiments of the invention can be implemented in conjunction with any other type of computing environment now known or developed hereafter.

[0106] Cloud computing is a service delivery model for enabling convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, network bandwidth, servers, processing, memory, storage, applications, virtual machines, and services) that can be rapidly provisioned and released with minimal management effort or interaction with service providers. This cloud model may include at least five features, at least three service models, and at least four deployment models.

[0107] The characteristics are as follows:

[0108] On-demand self-service: Cloud consumers can unilaterally and automatically provide computing power, such as server time and network storage, as needed, without requiring manual interaction with the service provider.

[0109] Wide Area Network (WAN) Access: Capabilities are available on the network and accessed through standard mechanisms that facilitate the use of heterogeneous thin or thick client platforms (e.g., mobile phones, laptops, and PDAs).

[0110] Resource pooling: A provider's computing resources are pooled to serve multiple consumers using a multi-tenant model, where different physical and virtual resources are dynamically allocated and reallocated based on demand. Location independence has significance because consumers typically do not control or know the exact location of the resources provided, but can specify the location at a higher level of abstraction (e.g., country, state, or data center).

[0111] Rapid Flexibility: In some cases, the ability to scale outwards and inwards quickly and flexibly can be provided. For consumers, the available capacity often appears unlimited and can be purchased in any quantity at any time.

[0112] Measurement services: Cloud systems automatically control and optimize resource usage by leveraging metering capabilities at a level of abstraction appropriate to the service type (e.g., storage, processing, bandwidth, and active user accounts). Resource usage can be monitored, controlled, and reported, providing transparency to both the providers and consumers of the services being utilized.

[0113] The service model is as follows:

[0114] Software as a Service (SaaS): The capability offered to consumers is the ability to use the provider's applications running on cloud infrastructure. Applications can be accessed from various client devices through thin client interfaces such as web browsers (e.g., web-based email). Consumers do not manage or control the underlying cloud infrastructure, including the network, servers, operating system, storage, or even individual application capabilities, with possible exceptions such as limited user-specific application configuration settings.

[0115] Platform as a Service (PaaS): This provides consumers with the ability to deploy consumer-created or acquired applications onto cloud infrastructure using programming languages ​​and tools supported by the provider. Consumers do not manage or control the underlying cloud infrastructure, including networks, servers, operating systems, or storage, but they have control over the deployed applications and the configuration of any application hosting environments.

[0116] Infrastructure as a Service (IaaS): This provides consumers with the capability to deliver processing, storage, networking, and other basic computing resources that enable them to deploy and run arbitrary software, which may include operating systems and applications. Consumers do not manage or control the underlying cloud infrastructure, but they do have control over the operating system, storage, deployed applications, and possibly limited control over selected networking components (e.g., host firewalls).

[0117] The deployment model is as follows:

[0118] Private cloud: Cloud infrastructure operated solely by an organization. It can be managed by the organization or a third party and can exist on-site or off-site.

[0119] Community cloud: Cloud infrastructure shared by several organizations and supporting a specific community with shared concerns (e.g., tasks, security requirements, policies, and compliance considerations). It can be managed by an organization or a third party and can exist on-site or off-site.

[0120] Public cloud: Cloud infrastructure available to the general public or large industrial groups and owned by organizations that sell cloud services.

[0121] Hybrid cloud: A cloud infrastructure is a combination of two or more clouds (private, community, or public) that remain a single entity but are bound together by standardized or proprietary technologies that enable data and applications to be ported together (e.g., cloud bursting for load balancing between clouds).

[0122] Cloud computing environments are service-oriented, focusing on statelessness, loose coupling, modularity, and semantic interoperability. At the heart of cloud computing is the infrastructure of a network of interconnected nodes.

[0123] Now for reference Figure 9 The diagram illustrates an illustrative cloud computing environment 50. As shown, the cloud computing environment 50 includes one or more cloud computing nodes 10 to which local computing devices used by cloud consumers can communicate, such as personal digital assistants (PDAs) or cellular phones 54A, desktop computers 54B, laptop computers 54C, and / or automotive computer systems 54N. The nodes 10 can communicate with each other. They can be physically or virtually grouped (not shown) in one or more networks, such as private clouds, community clouds, public clouds, or hybrid clouds, or combinations thereof, as described above. This allows the cloud computing environment 50 to provide infrastructure, platform, and / or software as a service, without requiring cloud consumers to maintain resources on their local computing devices. It is understood that... Figure 9 The types of computing devices 54A-N shown are for illustrative purposes only, and computing node 10 and cloud computing environment 50 can communicate with any type of computerized device on any type of network and / or network-addressable connection (e.g., using a web browser).

[0124] Now for reference Figure 10 This demonstrates a cloud computing environment of 50 ( Figure 9 This provides a set of functional abstractions. These can be understood in advance. Figure 10 The components, layers, and functions shown are for illustrative purposes only, and embodiments of the invention are not limited thereto. As depicted, the following layers and corresponding functions are provided:

[0125] The hardware and software layer 60 includes hardware and software components. Examples of hardware components include: a host 61; a server 62 based on a RISC (Reduced Instruction Set Computer) architecture; a server 63; a blade server 64; a storage device 65; and a network and network components 66. In some embodiments, software components include network application server software 67 and database software 68.

[0126] The virtualization layer 70 provides an abstraction layer from which the following examples of virtual entities can be provided: virtual server 71; virtual storage 72; virtual network 73, including virtual private network; virtual application and operating system 74; and virtual client 75.

[0127] In one example, management layer 80 can provide the following functionalities: Resource Provisioning 81 provides dynamic procurement of computing resources and other resources used to perform tasks within the cloud computing environment. Metering and Pricing 82 provides cost tracking when utilizing resources in the cloud computing environment, as well as billing or invoicing for consuming these resources. In one example, these resources may include application software licenses. Security provides authentication for cloud consumers and tasks, and protection for data and other resources. User Portal 83 provides access to the cloud computing environment for consumers and system administrators. Service Level Management 84 provides cloud resource allocation and management to ensure that required service levels are met. Service Level Agreement (SLA) Planning and Fulfillment 85 provides pre-scheduling and procurement of cloud resources, where future needs are anticipated according to the SLA.

[0128] The workload layer 90 provides examples of functionalities that can be leveraged in a cloud computing environment. Examples of workloads and functionalities that can be provided from this layer include: mapping and navigation 91; software development and lifecycle management 92; virtual classroom education delivery 93; data analytics and processing 94; transaction processing 95; and non-linear performance degradation prediction 96.

[0129] Embodiments of the present invention can be systems, methods, and / or computer program products at any possible level of technical detail integration. A computer program product may include a computer-readable storage medium (or media) having computer-readable program instructions thereon for causing a processor to perform aspects of the present invention.

[0130] Computer-readable storage media can be tangible devices capable of retaining and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of computer-readable storage media includes the following: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable optical disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices such as punch cards or recessed structures with instructions recorded thereon, and any suitable combination of the foregoing. As used herein, computer-readable storage media should not be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0131] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to a suitable computing / processing device, or via a network, such as the Internet, a local area network (LAN), a wide area network (WAN), and / or a wireless network, to an external computer or external storage device. The network may include copper cables, optical fibers, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to a computer-readable storage medium within the respective computing / processing device.

[0132] Computer-readable program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages ​​(including object-oriented programming languages ​​such as Smalltalk, C++, etc.) and procedural programming languages ​​(such as the "C" programming language or similar programming languages). The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, to perform aspects of this invention, electronic circuits, including, for example, programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), may execute computer-readable program instructions to personalize the electronic circuits by utilizing state information from the computer-readable program instructions.

[0133] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0134] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / actions specified in one or more blocks of a flowchart and / or block diagram. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and / or other device to operate in a particular manner, such that the computer-readable storage medium in which the instructions are stored includes an article of writing comprising instructions for implementing aspects of the functions / actions specified in one or more blocks of a flowchart and / or block diagram.

[0135] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer-implemented process, such that the instructions, which execute on the computer, other programmable apparatus or other device, perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0136] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or subset of instructions comprising one or more executable instructions for implementing a specified logical function. In some alternative embodiments, the functions indicated in the blocks may occur in a non-consecutive order as shown in the figures. For example, two blocks shown consecutively may actually be executed substantially simultaneously, or these blocks may sometimes be executed in reverse order, depending on the functions involved. It will also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified function or action or executes a combination of dedicated hardware and computer instructions.

[0137] While it is understandable that processing software (e.g., stored on) can be loaded via storage media such as CDs or DVDs, Figure 8 Any instruction in instruction 860 and / or configured to perform about Figure 3 -5 Any part and / or implementation of the method described in -5 Figure 1-2The processing software can be deployed by manually loading it directly onto the client, server, and agent computers (and / or any part of the functionality discussed in 6-7), but it can also be deployed automatically or semi-automatically by sending the processing software to a central server or a set of central servers. The processing software is then downloaded to the client computer that will execute it. Alternatively, the processing software can be sent directly to the client system via email. The processing software is then extracted into or loaded into a directory by executing a set of program instructions. Another alternative is to send the processing software directly to a directory on the client computer's hard drive. When a proxy server is present, the process selects a proxy server code, determines which computers the proxy server code will be placed on, transmits the proxy server code, and then installs the proxy server code on the agent computers. The processing software is then sent to the proxy server, where it is stored.

[0138] Embodiments of the present invention can also be delivered as part of a service agreement with client companies, non-profit organizations, government entities, internal organizational structures, etc. These embodiments may include configuring computer systems to perform, and deploying some or all of the software, hardware, and web services that implement some or all of the methods described herein. These embodiments may also include analyzing client operations, creating recommendations in response to the analysis, building systems that implement subsets of the recommendations, integrating the systems into existing processes and infrastructure, metering system usage, allocating fees to users of the system, and invoicing, issuing invoices (e.g., generating invoices), or otherwise receiving payment for the use of the system.

[0139] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the various embodiments. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the terms “comprising” and / or “including”, when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. In the preceding detailed description of exemplary embodiments of various embodiments, reference has been made to the accompanying drawings (in which like reference numerals denote like elements), which form part of the invention, and in which specific exemplary embodiments in which various embodiments may be practiced are illustrated. These embodiments have been described in sufficient detail to enable those skilled in the art to practice them, but other embodiments may also be used, and logical, mechanical, electrical, and other changes may be made without departing from the scope of the various embodiments. In the preceding description, numerous specific details have been set forth to provide a thorough understanding of the various embodiments. However, various embodiments may be practiced without these specific details. In other instances, well-known circuits, structures, and techniques have not been shown in detail so as not to obscure the embodiments.

[0140] Different instances of the term "embodiment" used in this specification do not necessarily refer to the same embodiment, but they may refer to the same embodiment. Any data and data structures shown or described herein are merely examples, and in other embodiments, different amounts of data, data types, the number and types of fields, field names, the number and types of rows, records, entries, or organization of data may be used. Furthermore, any data can be logically combined, thus eliminating the need for a separate data structure. Therefore, the foregoing detailed description should not be construed as limiting.

[0141] Various embodiments of this disclosure have been described for illustrative purposes, but are not intended to be exhaustive or limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein has been chosen to best explain the principles of the embodiments, their practical application, or improvements to existing technologies in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

[0142] Although this disclosure has been described with reference to specific embodiments, changes and modifications thereto will be apparent to those skilled in the art. Therefore, the appended claims are intended to be construed as covering all such alterations and modifications that fall within the true spirit and scope of this disclosure.

[0143] Any advantages discussed in this disclosure are exemplary advantages, and embodiments of this disclosure may exist that implement all, some, or none of the advantages discussed, while remaining within the spirit and scope of this disclosure.

[0144] The following provides a list of non-limiting embodiments to illustrate some aspects of this disclosure. Example 1 is a computer-implemented method. A new data sample is input into a fault prediction model, wherein the fault prediction model is trained using a labeled historical dataset, wherein corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples, wherein the corresponding training samples are clustered into multiple clusters, and wherein each of the multiple clusters is associated with a normal score and an anomaly score; and a classification associated with the new data sample is output based on a comparison of a first anomaly score of a first cluster including the new data sample among the multiple clusters with a normal score greater than the anomaly score among the multiple clusters.

[0145] Example 2 includes the approach of Example 1, but includes or excludes optional features. In this example, the classification indicates the likelihood of wear-related performance degradation of the asset associated with the new data sample.

[0146] Example 3 includes the methods of any of Examples 1 to 2, including or excluding optional features. In this example, the review window defines the number of sequentially first data points to be included in each corresponding training sample.

[0147] Example 4 includes the method of any of Examples 1 to 3, including or excluding optional features. In this example, the prediction time domain defines a predefined amount of future time, and wherein the corresponding label of the corresponding data point of the predefined amount of future time is associated with the corresponding training sample.

[0148] Example 5 includes the methods of any of Examples 1 through 4, including or excluding optional features. In this example, K-means clustering is used to cluster the corresponding training samples.

[0149] Example 6 includes the method of any one of Examples 1 to 5, including or excluding optional features. In this example, the method is performed by a fault prediction system based on software downloaded to the fault prediction system from a remote data processing system. Optionally, the method further includes: the use of metering software; and generating invoices based on metering usage.

[0150] Example 7 is a system. The system includes one or more processors; and one or more computer-readable storage media storing program instructions, which, when executed by the one or more processors, are configured to cause the one or more processors to perform a method comprising: inputting a new data sample into a fault prediction model, wherein the fault prediction model is trained using a labeled historical dataset, wherein corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples, wherein the corresponding training samples are clustered into multiple clusters, and wherein each of the multiple clusters is associated with a normal score and an anomaly score; and outputting a classification associated with the new data sample based on a comparison of a first anomaly score of a first cluster including the new data sample among the multiple clusters with a normal score greater than the anomaly score among the multiple clusters.

[0151] Example 8 includes the system of Example 7, including or excluding optional features. In this example, the classification indicates the likelihood of wear-related performance degradation of the asset associated with the new data sample.

[0152] Example 9 includes a system that incorporates or excludes optional features from any of Examples 7 to 8. In this example, the review window defines the number of sequentially preceding data points to be included in each corresponding training sample.

[0153] Example 10 includes a system comprising any of Examples 7 through 9, including or excluding optional features. In this example, the prediction time domain defines a predefined amount of time in the future, and wherein the corresponding label of the corresponding data point for the predefined amount of time in the future is associated with the corresponding training sample.

[0154] Example 11 includes a system from any of Examples 7 through 10, including or excluding optional features. In this example, K-means clustering is used to cluster the corresponding training samples.

[0155] Example 12 is a computer program product. The computer program product includes one or more computer-readable storage media and program instructions commonly stored on the one or more computer-readable storage media. The program instructions include instructions configured to cause one or more processors to perform a method comprising: inputting a new data sample into a fault prediction model; training the fault prediction model using a labeled historical dataset; associating corresponding data points with a review window and a prediction time domain to create corresponding training samples; clustering the corresponding training samples into multiple clusters; and associating each of the multiple clusters with a normal score and an anomaly score; and outputting a classification associated with the new data sample based on a comparison of a first anomaly score of a first cluster including the new data sample among the multiple clusters with a normal score greater than the anomaly score among the multiple clusters.

[0156] Example 13 includes the computer program product of Example 12, including or excluding optional features. In this example, the classification indicates the likelihood of wear-related performance degradation of the asset associated with the new data sample.

[0157] Example 14 includes a computer program product of any of Examples 12 to 13, including or excluding optional features. In this example, the review window defines the number of sequentially preceding data points to be included in each corresponding training sample.

[0158] Example 15 includes a computer program product of any of Examples 12 through 14, including or excluding optional features. In this example, the prediction time domain defines a predefined amount of time in the future, and wherein a corresponding label for a data point of the predefined amount of time in the future is associated with a corresponding training sample.

[0159] Example 16 includes a computer program product of any of Examples 12 through 15, including or excluding optional features. In this example, K-means clustering is used to cluster the corresponding training samples.

[0160] Example 17 is a computer-implemented method for predicting wear-related degradation of a screw pump (PCP). The method includes inputting a new data sample of PCP into a model configured to predict wear-related degradation of PCP, wherein the model is trained using a labeled historical PCP dataset, wherein corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples, wherein the corresponding training samples are clustered into multiple clusters, and wherein each of the multiple clusters is associated with a normal score and an anomaly score; and outputting a classification associated with the new data sample based on a comparison of a first anomaly score of a first cluster including the new data sample in the multiple clusters with a normal score greater than the anomaly score in the multiple clusters, wherein the classification indicates wear-related degradation of the PCP.

[0161] Example 18 is a computer-implemented method for predicting wear-related degradation of a screw pump (PCP), comprising: generating labeled historical data by performing binary labeling of historical data associated with one or more PCPs; generating multiple training data samples by applying a review window and a prediction time domain to the corresponding data points of the labeled historical data; clustering the multiple training data samples into multiple clusters; calculating a cluster score for the corresponding clusters in the multiple clusters; assigning a new data sample of PCP to a first cluster in the multiple clusters; and assigning a classification to the new data sample based on the cluster score associated with the first cluster, wherein the classification indicates the likelihood of future wear-related degradation of the PCP.

[0162] Example 19 includes the method of Example 18, including or excluding optional features. In this example, historical data is marked as faulty within a predetermined time period prior to a known pump replacement date.

[0163] Example 20 includes the method of any of Examples 18 to 19, including or excluding optional features. In this example, the tagged historical data includes pump speed data, pump torque data, casing pressure data, productivity data, and maintenance records.

[0164] Example 21 includes the method of any one of Examples 18 to 20, including or excluding optional features. In this example, calculating the cluster score of the corresponding cluster further includes: calculating a normal score for a first cluster, wherein the normal score is a first proportion of training data samples associated with normal states in the first cluster divided by a second proportion of training data samples associated with normal states in a plurality of training data samples; and calculating an anomaly score for the first cluster, wherein the anomaly score is a third proportion of training data samples associated with deteriorated states in the first cluster divided by a fourth proportion of training data samples associated with deteriorated states in a plurality of training data samples. Optionally, the classification is based on the larger of the normal score or the anomaly score of the first cluster.

[0165] Example 22 includes the method of any one of Examples 18 to 21, including or excluding optional features. In this example, the method includes generating a fault signal for a new data sample, wherein the fault signal includes the average anomaly score of the new data sample over a predetermined number of previous data points. Optionally, generating the fault signal further includes: calculating the average anomaly score of clusters with normal scores greater than the anomaly score in a plurality of clusters; for each of the predetermined number of previous data points, associating a value with an anomaly score in a first cluster that is greater than the average anomaly score, and associating a zero value with an anomaly score in the first cluster that is less than the average anomaly score; and calculating the fault signal as the average of the value and the zero value associated with each of the predetermined number of previous data points.

Claims

1. A computer-implemented method for predicting performance degradation, comprising: New data samples based on sensor data are input into a fault prediction model, wherein the fault prediction model is trained using a labeled historical dataset, wherein corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples, wherein the corresponding training samples are clustered into multiple clusters, and wherein each of the multiple clusters is associated with a normal score and an abnormal score; and Based on comparing the first anomaly score of the first cluster, which includes the new data sample, with the average anomaly score of the clusters that have a normal score greater than the anomaly score, a classification associated with the new data sample is output.

2. The method according to claim 1, wherein, The classification indicates the likelihood of wear-related performance degradation of assets associated with the new data sample.

3. The method according to claim 1, wherein, The review window definition should include the number of sequentially preceding data points in each corresponding training sample.

4. The method according to claim 1, wherein, The predicted time domain defines a predefined amount of time in the future, and wherein the corresponding labels of the corresponding data points of the predefined amount of time in the future are associated with the corresponding training samples.

5. The method according to claim 1, wherein, The corresponding training samples are clustered using K-means clustering.

6. The method according to claim 1, wherein, The method is executed by the fault prediction system based on software downloaded from the remote data processing system to the fault prediction system.

7. The method according to claim 6, wherein, The method further includes: Measuring the use of the software; and Invoices are generated based on the usage described in the measurement.

8. The method according to claim 1, wherein, The method is used to predict wear-related degradation of a screw pump (PCP), wherein the input step includes inputting new data samples of PCP into a model configured to predict wear-related degradation of the PCP, wherein the model is trained using a labeled historical PCP dataset, and wherein the classification output during the output step indicates wear-related degradation of the PCP.

9. The method of claim 8, comprising: Labeled historical data is generated by performing binary labeling of historical data associated with one or more PCPs, wherein multiple training data samples are created by applying a review window and a prediction time domain to the corresponding data points of the labeled historical data, wherein the method includes: Calculate a cluster score for the corresponding cluster among the plurality of clusters, wherein the cluster score is the normal score and the abnormal score; The new data samples from the PCP are allocated to the first cluster among the plurality of clusters; and The new data samples are assigned a classification based on the cluster score associated with the first cluster.

10. The method according to claim 9, wherein, The historical data of the marking is marked as faulty within a predetermined period of time prior to the known pump replacement date.

11. The method according to claim 9, wherein, The historical data for the markers includes pump speed data, pump torque data, casing pressure data, productivity data, and maintenance records.

12. The method according to claim 9, wherein, Calculating the cluster score of the corresponding cluster further includes: Calculate the normality score of the first cluster, wherein the normality score is a first proportion of training data samples in the first cluster associated with the normal state divided by a second proportion of training data samples in the plurality of training data samples associated with the normal state; and Calculate the anomaly score of the first cluster, wherein the anomaly score is the third proportion of the training data samples in the first cluster associated with the deteriorated state divided by the fourth proportion of the training data samples in the plurality of training data samples associated with the deteriorated state.

13. The method according to claim 12, wherein, The classification is based on the larger of the normal score or the abnormal score of the first cluster.

14. The method according to claim 9, further comprising: A fault signal is generated for the new data sample, wherein the fault signal includes the average anomaly score of the new data sample over a predetermined number of previous data points.

15. The method according to claim 14, wherein, Generating the fault signal also includes: Calculate the average abnormal score of the clusters among the plurality of clusters that have a normal score greater than the abnormal score; For each of the predetermined number of previous data points, a value is associated with a data point in the first cluster that has an anomaly score greater than the average anomaly score, and a zero value is associated with a data point in the first cluster that has an anomaly score less than the average anomaly score; and The fault signal is calculated as the average of the one value and the zero value associated with each of the predetermined number of previous data points.

16. A system for predicting performance degradation, comprising: One or more processors; as well as A computer-readable storage medium containing one or more stored program instructions, the program instructions being configured, when executed by the one or more processors, to cause the one or more processors to perform a method comprising: New data samples based on sensor data are input into a fault prediction model, wherein the fault prediction model is trained using a labeled historical dataset, wherein corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples, wherein the corresponding training samples are clustered into multiple clusters, and wherein each of the multiple clusters is associated with a normal score and an abnormal score; and Based on comparing the first anomaly score of the first cluster, which includes the new data sample, with the average anomaly score of the clusters that have a normal score greater than the anomaly score, a classification associated with the new data sample is output.

17. The system according to claim 16, wherein, The classification indicates the likelihood of wear-related performance degradation of assets associated with the new data sample.

18. The system according to claim 16, wherein, The review window definition should include the number of sequentially preceding data points in each corresponding training sample.

19. The system according to claim 16, wherein, The predicted time domain defines a predefined amount of time in the future, and wherein the corresponding label of the corresponding data point of the predefined amount of time in the future is associated with the corresponding training sample.

20. The system according to claim 16, wherein, The corresponding training samples are clustered using K-means clustering.

21. A computer program product comprising one or more computer-readable storage media and program instructions commonly stored on the one or more computer-readable storage media, the program instructions including instructions configured to cause one or more processors to perform a method for predicting performance degradation, the method comprising: New data samples based on sensor data are input into a fault prediction model, wherein the fault prediction model is trained using a labeled historical dataset, wherein corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples, wherein the corresponding training samples are clustered into multiple clusters, and wherein each of the multiple clusters is associated with a normal score and an abnormal score; and Based on comparing the first anomaly score of the first cluster, which includes the new data sample, with the average anomaly score of the clusters that have a normal score greater than the anomaly score, a classification associated with the new data sample is output.

22. The computer program product according to claim 21, wherein, The classification indicates the likelihood of wear-related performance degradation of the assets associated with the new data sample.

23. The computer program product according to claim 21, wherein, The review window definition should include the number of sequentially preceding data points in each corresponding training sample.

24. The computer program product according to claim 21, wherein, The predicted time domain defines a predefined amount of time in the future, and wherein the corresponding label of the corresponding data point of the predefined amount of time in the future is associated with the corresponding training sample.

25. The computer program product according to claim 21, wherein, The corresponding training samples are clustered using K-means clustering.

26. A computer-implemented method for predicting wear-related degradation of a screw pump (PCP), the method comprising: New PCP data samples are input into a model configured to predict wear-related degradation of the PCP, wherein the model is trained using a labeled historical PCP dataset, wherein corresponding data points are associated with a review window and a prediction time domain to create corresponding training samples, wherein the corresponding training samples are clustered into multiple clusters, and wherein each of the multiple clusters is associated with a normal score and an abnormal score; and Based on comparing the first anomaly score of the first cluster, which includes the new data sample, among the plurality of clusters with the average anomaly score of the clusters that has a normal score greater than the anomaly score, a classification associated with the new data sample is output, wherein the classification indicates the wear-related degradation of the PCP.

27. A computer-implemented method for predicting wear-related degradation of a screw pump (PCP), the method comprising: Historical data of tags is generated by performing binary tagging of historical data associated with one or more PCPs; Multiple training data samples are generated by applying a review window and a prediction time domain to the corresponding data points of the historical data of the labeled data; The multiple training data samples are clustered into multiple clusters; Calculate the cluster score of the corresponding cluster among the plurality of clusters; The new data sample from PCP is assigned to the first cluster among the plurality of clusters; as well as The new data samples are assigned a classification based on the cluster score associated with the first cluster, wherein the classification indicates the likelihood of future wear-related degradation of the PCP.

28. The method according to claim 27, wherein, The historical data of the marking is marked as faulty within a predetermined period of time prior to the known pump replacement date.

29. The method according to claim 27, wherein, The historical data for the markers includes pump speed data, pump torque data, casing pressure data, productivity data, and maintenance records.

30. The method according to claim 27, wherein, Calculating the cluster score of the corresponding cluster further includes: Calculate the normality score of the first cluster, wherein the normality score is a first proportion of training data samples in the first cluster associated with the normal state divided by a second proportion of training data samples in the plurality of training data samples associated with the normal state; and Calculate the anomaly score of the first cluster, wherein the anomaly score is the third proportion of the training data samples in the first cluster associated with the deteriorated state divided by the fourth proportion of the training data samples in the plurality of training data samples associated with the deteriorated state.

31. The method according to claim 30, wherein, The classification is based on the larger of the normal score or the abnormal score of the first cluster.

32. The method of claim 27, further comprising: A fault signal is generated for the new data sample, wherein the fault signal includes the average anomaly score of the new data sample over a predetermined number of previous data points.

33. The method according to claim 32, wherein, Generating the fault signal also includes: Calculate the average abnormal score of the clusters among the plurality of clusters that have a normal score greater than the abnormal score; For each of the predetermined number of previous data points, a value is associated with a data point in the first cluster that has an anomaly score greater than the average anomaly score, and a zero value is associated with a data point in the first cluster that has an anomaly score less than the average anomaly score; and The fault signal is calculated as the average of the one value and the zero value associated with each of the predetermined number of previous data points.

34. A computer program product comprising program code means, wherein when the program is run on a computer, the program code means is adapted to perform the method according to any one of claims 1 to 15 and 26 to 33.

Images