Process-data processing method and computer device

Abstract

Provided in the present application are a process-data processing method and a computer device, which can be applied to the fault determination of a device. The method comprises: on the basis of n types of historical process data of m devices, obtaining first process data, and performing slow feature transformation to obtain a first transformation matrix; on the basis of second process data, periodically updating the first transformation matrix to obtain a second edge transformation matrix, wherein the second process data is obtained on the basis of periodically collected process data on a target device; and when the value of real-time target process data of the target device is within a first range, and on the basis of the second transformation matrix and a baseline value, performing fault determination on the target device. In the present application, the deep mining of slow feature quantities which describe the essential state of a target device is realized by integrating multi-dimensional process data, thereby achieving strong interpretability; by means of performing iterative incremental updating of a first transformation matrix on a target device, the burden of calculation and storage of the target device during updating is reduced; and when determination is performed on the target device, the actual situation of an operation condition is taken into account, thereby eliminating false alarms.

Description

A method for processing process data and a computer device

[0001] This application claims priority to Chinese Patent Application No. 202510072031.6, filed on January 15, 2025, entitled "A Method for Processing Process Data and a Computer Device", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of data processing, and more particularly to a method for processing process data and a computer device. Background Technology

[0003] With the rapid development of industrial automation and intelligent manufacturing, massive amounts of process data generated during production are accumulating. This data covers information such as equipment operating status, process parameters, and environmental conditions, and can be used to reveal potential failure modes and trends. Compared to traditional vibration methods that rely on complex sensors and signal processing technologies, process data offers significant advantages in terms of real-time performance, adaptability, and cost-effectiveness. By utilizing data generated during production (such as temperature, pressure, flow rate, and energy consumption) combined with data analysis techniques, comprehensive equipment status monitoring and fault prediction can be achieved with virtually no additional cost.

[0004] Currently, the fault prediction mechanism used in the industry is an alarm strategy based on mechanistic thresholds. Its implementation involves setting a threshold for a single-dimensional variable (e.g., flow rate), typically with two alarm levels: "high alarm" and "high-high alarm." When the real-time monitored value exceeds the threshold, an alarm is generated, reminding workers to check for faults; otherwise, it is considered normal.

[0005] However, the above alarm strategy has the following drawbacks: 1. High false negative rate: This is because the signal-to-noise ratio of process data is low, and the information provided by single-dimensional data is limited, thus making it easy to produce false negatives; 2. Poor generalization: Due to the significant differences between different devices, it is difficult for single signal threshold alarms to achieve threshold self-adaptation. Usually, parameters need to be configured for each device, which is a lot of work and results in poor generalization; 3. High false alarm rate: Single-dimensional data cannot obtain the device operating conditions through other dimension information (such as current, speed, etc.) to update the threshold in real time, so it is easy to produce false alarms when affected by the external environment. Summary of the Invention

[0006] This application provides a process data processing method and a computer device for in-depth mining of slow feature quantities describing the essential state of a target device by combining multi-dimensional process data (i.e., n types). It has strong interpretability. By iteratively and incrementally updating the first transformation matrix on the target device, the computational and storage burden of the target device during the update is reduced. Finally, when judging the target device, the actual situation of the operating condition indication quantity (i.e., real-time target process data) is taken into account to eliminate false alarms.

[0007] Based on this, the embodiments of this application provide the following technical solutions:

[0008] Firstly, this application provides a method for processing process data. This method specifically includes: firstly, acquiring historical process data of n types from m devices; then, obtaining first process data based on these n types of historical process data, where m ≥ 1 and n ≥ 2. It should be noted that the m devices can be different types of devices with similar functions, and the n types refer to the types of process data generated by the m devices with similar functions. Next, a slow feature transformation is performed on the obtained first process data to obtain a transformation matrix of the slow features, which can be called the first transformation matrix. It should be noted that the first process data is data fused from the n types of process data, thereby achieving dimensionality reduction representation of multi-dimensional data. This first process data can be considered an initial matrix. The process of performing a slow feature transformation on the first process data is essentially: initial matrix * first transformation matrix = slow feature. The first transformation matrix contains the core key features of the initial matrix. Then, the obtained first transformation matrix is ​​deployed on the target device. Subsequently, the first transformation matrix can be periodically updated based on the second process data. This periodic update can be an incremental iterative update until the update termination condition is met. The process data on the target device includes the aforementioned n types of data. The second process data is obtained based on the periodically collected process data on the target device. Finally, given that the real-time target process data on the running target device falls within a first range, a fault determination is performed on the running target device based on the second transformation matrix and the baseline value. The data type of the target process data is a target type, which is one of the n types. The second transformation matrix is ​​the first transformation matrix obtained from the last update, and the baseline value is used to characterize the alarm boundary.

[0009] In the above embodiments of this application, variable contribution information (i.e., first process data) is extracted from the prior knowledge (i.e., n types of historical process data) of multiple similar devices (i.e., m devices, such as m pumps with different functions). Multi-dimensional joint analysis of different types of data across multiple dimensions is then performed to eliminate interference from random processes such as operating conditions and processes, thereby enhancing the slow feature quantities that describe the essential state of the equipment and improving interpretability. Secondly, a first transformation matrix is ​​trained using the historical process data of all m devices. Then, the first transformation matrix is ​​iteratively and incrementally updated (i.e., periodically updated) on the target device to reduce the computational and storage burden of updating the transformation matrix, while maintaining high accuracy. Finally, when judging the target device, the operating condition indication (i.e., real-time target process data) is considered to eliminate false alarms caused by start-up / shutdown / operating condition fluctuations, i.e., false alarms are reduced through a double-check method. Therefore, the above embodiments of this application address the fault diagnosis problem of specific equipment (i.e., the target device) from three aspects: feature enhancement (solving missed alarms), transformation matrix update (improving generalization), and double-check (solving false alarms).

[0010] In one possible implementation of the first aspect, the second process data includes process data for the current period and process data for historical periods, where the historical periods are those preceding the current period. One implementation for periodically updating the first transformation matrix based on the second process data could be as follows: First, obtain the current transformation matrix increment based on the current period's process data. This current transformation matrix increment characterizes the influence of the current period's process data on the first transformation matrix updated in the previous period, where the previous period refers to the period preceding the current period; for example, the period preceding period t2 is t1. Then, update the first transformation matrix updated in the previous period based on the current transformation matrix increment, thereby obtaining the updated first transformation matrix for the current period. This operation is performed for each current period until the update termination condition is met.

[0011] In the above embodiments of this application, the first transformation matrix of the previous update is updated based on the current transformation matrix increment of each cycle, thereby deriving the matrix transformation law and obtaining the analytical solution of the iterative update of the statistical quantity (i.e. the process data of each cycle) increment between different batches, realizing the lossless update of the first transformation matrix, and thus greatly reducing the computation and storage requirements when updating the first transformation matrix to meet the deployment requirements on the edge side.

[0012] In one possible implementation of the first aspect, the baseline value is obtained based on the second transformation matrix and the non-fault history process data on the target device.

[0013] In the above embodiments of this application, each target device can calculate its own unique baseline value, and different target devices can calculate different baseline values, thereby realizing a flexible fault determination method.

[0014] In one possible implementation of the first aspect, obtaining the first process data based on n types of historical process data from m devices can be achieved as follows: First, acquire n types of historical process data from the aforementioned m devices during a first historical time period (e.g., the past week). This first historical time period can be customized based on actual needs, and this application does not impose any limitations on it. Then, determine at least one sub-time period (assuming k sub-time periods) from this first historical time period. This sub-time period refers to the period when the equipment operates stably. This is because during equipment operation, start-up and shutdown, external environmental influences, etc., can affect the equipment's operating condition. Therefore, it is necessary to select historical process data from the first historical time period during the stable operating period for subsequent processing, thereby initially eliminating some random interference from operating conditions, processes, etc. Assuming that the historical process data included in each sub-time period is called stable process data, the first process data can finally be obtained based on the historical process data (i.e., stable process data) included in each selected sub-time period.

[0015] In the above embodiments of this application, since the equipment's operating conditions are affected by start-up and shutdown, external environmental influences, etc., during the operation of the equipment, historical process data of the stable operating period of the equipment is selected from the first historical period for subsequent processing, thereby initially eliminating some random interferences such as operating conditions and processes, and improving the accuracy of subsequent fault determination.

[0016] In one possible implementation of the first aspect, determining at least one sub-period from the first historical period can be achieved by: firstly, determining one type (which may be called the first type) from n types, and then determining the at least one sub-period from the first historical period using historical process data belonging to the first type.

[0017] In the above embodiments of this application, it is specifically described that the purpose of determining the sub-period from the first historical period is achieved by using process data belonging to the first type, which is feasible.

[0018] In one possible implementation of the first aspect, the equipment condition can be identified by performing an Augmented Dickey-Fuller test (ADF) on historical process data belonging to the first type, thereby determining the at least one sub-period from the first historical period.

[0019] In the above embodiments of this application, the ADF method is used to identify the equipment operating conditions to determine the sub-time period, which is simple and easy to operate.

[0020] In one possible implementation of the first aspect, obtaining the first process data based on historical process data in at least one sub-period could be achieved by performing a difference operation on the historical process data included in each sub-period (all sub-periods of m devices), thereby obtaining the time series information within each segment of stationary process data. Thus, each sub-period corresponds to a segment of stationary process data and a segment of time series information. All stationary process data and all time series information across all sub-periods constitute the aforementioned first process data. It should be noted that, assuming the historical process data included in a certain sub-period consists of p target data points, then performing a difference operation on the historical process data included in that sub-period refers to performing a difference operation on those p specific target data points, with the aim of establishing the temporal correlation between the various target data points within that sub-period.

[0021] In the above embodiments of this application, by performing differential operations on the process data of each sub-time period, the temporal correlation of each target data within each process data segment is obtained, thereby improving the accuracy of subsequent slow feature calculation.

[0022] In one possible implementation of the first aspect, one way to obtain the first transformation matrix of the slow features by performing a slow feature transformation on the first process data is as follows: First, perform an exponential basis transformation on the first process data to obtain a second matrix; then, perform a slow feature transformation on the second matrix to obtain the first transformation matrix of the slow features.

[0023] In the above embodiments of this application, the first process data is first subjected to an exponential basis transformation to obtain a second matrix, and then a slow feature transformation is performed on the second matrix to obtain the first transformation matrix. This results in more discriminative data contained in the slow features, further improving the sensitivity of the first transformation matrix.

[0024] In one possible implementation of the first aspect, the method may further include: if the value of the real-time target process data (also referred to as the operating condition indicator variable) on the operating target equipment is not within a first range, then the fault determination of the target equipment shall be stopped. That is, if the operating condition indicator variable indicates shutdown or non-stationarity (i.e., the value of the operating condition indicator variable falls within other value ranges representing shutdown or non-stationarity), then the alarm shall be canceled.

[0025] In the above embodiments of this application, if the real-time target process data (i.e., the operating condition indicator variable) is outside the first range (i.e., indicating shutdown or non-stationarity), the alarm is canceled and no fault determination is performed. This is because due to the complexity of the operating conditions, statistical quantities are easily subject to temporary over-limit situations caused by external influences, thus generating false alarms. Therefore, alarms when the operating condition indicator target equipment is running stably are valid alarms, thereby greatly improving alarm accuracy.

[0026] In one possible implementation of the first aspect, the conditions for terminating the update include, but are not limited to: 1) the first transformation matrix obtained from the last update converges; 2) a preset update duration is reached; and 3) a preset update round is reached.

[0027] In the above embodiments of this application, several implementation methods for achieving the update termination condition are specifically described, which have wide applicability and flexibility.

[0028] A second aspect of this application provides a computer device having the function of implementing the method of the first aspect or any possible implementation thereof. This function can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above-described function.

[0029] A third aspect of this application provides a computer device that may include a memory, a processor, and a bus system. The memory is used to store a computer program (also referred to as a program or computer-readable instructions), and the processor is used to invoke the program stored in the memory to execute the method of the first aspect of the embodiments of this application or any possible implementation of the first aspect.

[0030] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a computer, enable the computer to perform the methods described in the first aspect or any possible implementation thereof.

[0031] The fifth aspect of this application provides a computer program or a computer program product containing instructions that, when the computer program or computer program product is run on a computer, causes the computer to perform the method described in the first aspect or any possible implementation of the first aspect.

[0032] A sixth aspect of this application provides a chip including at least one processor and at least one interface circuit coupled to the processor. The interface circuit performs transceiver functions and sends instructions to the at least one processor. The at least one processor runs a computer program or instructions, having the functionality to implement the methods described in the first aspect or any possible implementation of the first aspect. This functionality can be implemented in hardware, software, or a combination of hardware and software, including one or more modules corresponding to the described functions. Furthermore, the interface circuit is used to communicate with other modules outside the chip.

[0033] In some implementations of this application, some of the one or more processors may implement some steps of the above method through dedicated hardware. For example, the processing involving neural network models may be implemented by a dedicated neural network processor or graphics processor.

[0034] The method provided in this application embodiment can be implemented by a single chip or by multiple chips working together. Attached Figure Description

[0035] Figure 1 is a schematic diagram of the system architecture provided in an embodiment of this application;

[0036] Figure 2 is a flowchart illustrating a process data processing method provided in an embodiment of this application.

[0037] Figure 3 is a schematic diagram of an incremental update method provided in an embodiment of this application;

[0038] Figure 4 is a flowchart illustrating a process data processing method provided in an embodiment of this application.

[0039] Figure 5 is a flowchart of a traditional fault diagnosis method based on process control signals;

[0040] Figure 6 is a schematic diagram of a traditional retraining process;

[0041] Figure 7 is a schematic diagram of a traditional alarm post-processing flow;

[0042] Figure 8 is a schematic diagram of an overall process provided in an embodiment of this application;

[0043] Figure 9 is a schematic diagram of a computer device provided in an embodiment of this application;

[0044] Figure 10 is a schematic diagram of a computer device provided in an embodiment of this application. Detailed Implementation

[0045] This application provides a process data processing method and computer equipment, used to extract variable contribution information (i.e., first process data) from prior knowledge (i.e., n types of historical process data) of multiple similar devices (i.e., m devices, such as m pumps with different functions), and perform multi-dimensional joint analysis to eliminate random process interference such as operating conditions and processes, thereby achieving in-depth mining of slow feature quantities describing the essential state of the equipment; secondly, a first transformation matrix is ​​obtained by training with all m devices' historical process data, and then the first transformation matrix is ​​iteratively and incrementally updated (i.e., periodically updated) on the target device to reduce the computational and storage burden of the device when updating the transformation matrix; finally, when judging the target device, the operating condition indication (i.e., real-time target process data) is taken into account to eliminate false alarms caused by start-up / shutdown / operating condition fluctuations. Thus, from three aspects—feature enhancement (solving missed alarms), matrix update (improving generalization), and double verification (solving false alarms)—the fault diagnosis problem of specific equipment (i.e., target equipment) is addressed in a targeted manner.

[0046] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0047] To better understand the solutions of the embodiments of this application, the relevant terms and concepts that may be involved in the embodiments of this application will be introduced below. It should be understood that the explanation of the relevant concepts may be limited due to the specific circumstances of the embodiments of this application, but it does not mean that this application can only be limited to that specific situation. The specific circumstances of different embodiments may also differ, and no specific limitation is made here.

[0048] (1) Process data

[0049] Process data, also known as production process data, refers to the data generated by equipment during operation that characterizes the equipment's operating status (such as speed, flow rate, etc.), process parameters (such as RMS current, pressure, temperature, etc.), and environmental conditions (such as room temperature, humidity, etc.). Taking a pump as an example, the process data generated during its operation can include real-time data such as RMS current, flow rate, and temperature.

[0050] (2) Slow Feature Analysis (SFA)

[0051] Slow Feature Extraction (SFA) is an unsupervised learning method, also known as Slow Feature Transform, designed to extract important, slowly changing features from high-dimensional time series data. It is primarily used to discover temporal evolution patterns and dynamic features in data, and is commonly applied in time series forecasting, robot motion planning, and other fields. The goal of SFA is to learn a transformation function (also known as a transformation matrix) to map the input data to a new representation space, in which important slow features change relatively slowly over time.

[0052] (3) Transformation matrix

[0053] A transformation matrix is ​​a concept in linear algebra. In linear algebra, linear transformations can be represented using matrices. If T is a linear transformation that maps Rn to Rm, and x is a column vector with n elements, then we call the m×n matrix A the transformation matrix of T. Any linear transformation can be represented by matrices in a computationally pleasing and consistent form, and multiple transformations can be easily concatenated together by matrix multiplication.

[0054] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0055] First, the system architecture and overall process of the method used in the embodiments of this application will be described. Please refer to Figure 1 for details. Figure 1 is a schematic diagram of the system architecture provided in the embodiments of this application. The system architecture includes a central cloud 101, edge nodes 102, and end-side devices 103. The central cloud 101 has a predictive maintenance platform for equipment, which is used to realize functions such as anomaly warning and intelligent operation and maintenance. The central cloud 101 and the edge nodes 102 (such as computer equipment, mobile phones, etc.) are connected through a network (such as a park intranet). The edge nodes 102 are used to realize functions such as data acquisition and related calculation to obtain calculation results. The end-side devices 103 are devices that generate process data. They can be key rotating equipment in industries such as chemical and energy (such as fans, pumps, etc.), or automatic production equipment on production lines (such as filling machines, welding machines, etc.), or some sensors on the equipment (such as temperature sensors, counters, etc.). Specifically, this application does not limit the type of end-side devices 103, but determines it according to the actual application scenario.

[0056] It should be noted that, in the embodiments of this application, the system architecture in Figure 1 is only illustrative, and there are no limitations on the deployment method of each unit.

[0057] Based on the above system architecture, the process data processing method provided in the embodiments of this application will be described below. For ease of understanding, the edge node is taken as a computer device as an example. Please refer to Figure 2 for details. Figure 2 is a flowchart of the process data processing method provided in the embodiments of this application, which may include the following steps:

[0058] 201. Obtain the first process data based on the historical process data of m devices of n types, where m≥1 and n≥2.

[0059] First, the computer device acquires historical process data of n types from m devices, and then obtains the first process data based on the historical process data of n types, where m≥1 and n≥2.

[0060] It should be noted that, in the embodiments of this application, the m devices can be different types of devices with similar functions. As an example, a pump is a machine that transports or pressurizes fluids. Therefore, all pumps have the similar function of transporting or pressurizing fluids, but the types of pumps can be diverse. For example, the types of pumps include, but are not limited to: centrifugal pumps, axial flow pumps, mixed flow pumps, vortex pumps, gear pumps, vane pumps, screw pumps, etc.

[0061] It should also be noted that in some embodiments of this application, "n types" refers to the types of process data generated by m devices with similar functions. Generally, devices with similar functions generate the same types of process data during production and operation. Taking the pump mentioned above as an example, different types of pumps will generate process data including: RMS current, flow rate, temperature, etc.

[0062] It should also be noted that in some embodiments of this application, all types of process data generated by the device can be used as historical process data for subsequent processing. For example, assuming a total of 5 types of process data are generated, all 5 types of process data can be used for subsequent processing (i.e., n=5). Alternatively, several types of process data can be selected from all types as historical process data for subsequent processing. For example, assuming a total of 5 types of process data are generated, only the 3 most important types of process data can be selected for subsequent processing (i.e., n=3). This application does not limit the specifics of this.

[0063] It should be noted that in some embodiments of this application, the method of obtaining the first process data based on n types of historical process data can be as follows: First, obtain the n types of historical process data of the above-mentioned m devices in a historical first time period (e.g., the past week). This historical first time period can be customized based on actual needs, and this application does not limit it. Then, determine at least one sub-time period (assuming k sub-time periods) from this historical first time period. This sub-time period refers to the period when the equipment is operating stably. This is because during the operation of the equipment, start-up and shutdown, external environmental influences, etc., will affect the operating condition of the equipment. Therefore, it is necessary to select the historical process data of the stable operating period of the equipment from the historical first time period for subsequent processing, thereby initially eliminating some random interferences such as operating conditions and processes. It should be noted that in some embodiments of this application, the sub-time period can be determined based on the selected operating condition indicator variable, that is, one type (which can be called the first type) of process data can be determined from the n types to achieve the purpose of determining the sub-time period from the historical first time period. Based on this, the stable operating period of the equipment can be identified, and the data type that best reflects the operating condition of the equipment can be selected. Assuming that the historical process data included in each sub-period is called stationary process data, the first process data can be obtained based on the historical process data (i.e., stationary process data) included in each selected sub-period.

[0064] It should also be noted that, in some embodiments of this application, the equipment operating condition can be identified by performing ADF on historical process data belonging to the first type, thereby determining the at least one sub-period from the first historical period. Specifically, using the ADF algorithm, if a sub-period in the first historical period passes the test, it is identified as a stable period; if the sub-period fails the test, it is identified as a non-stable period. Based on this, each stable operating segment of each device in the first historical period can be found.

[0065] To facilitate understanding of the above process, a specific example will be used below:

[0066] Taking m devices as m chemical industrial pumps as an example, the corresponding historical process data is assumed to include three types: RMS current, flow rate, and temperature. First, assuming the past week is the first historical period, the RMS current data, flow rate data, and temperature data collected periodically over the past week (e.g., assuming RMS current, flow rate, and temperature are collected every minute for the m pumps) constitute the n types of historical process data for the m devices in the first historical period.

[0067] Next, taking device a as an example, suppose device a had three stable operating periods in the past week (let's say days 1-2, day 4, and day 6-7). These three periods can be identified using the RMS current value as an indicator variable. For example, the RMS current value can be used to determine whether the pump is running and operating normally. Specifically, this can be achieved by applying a stability discrimination method (such as ADF) to the RMS current value to identify the pump's operating condition. It's important to note that for other types of equipment, such as internal combustion engines, the engine's load can be characterized using data like engine speed, thus identifying the stable operating periods. Different equipment uses different data types as indicator variables, which will not be elaborated upon here.

[0068] Therefore, from the historical process data A of device a over the past week, the stable process data A1 corresponding to days 1-2, A2 corresponding to day 4, and A3 corresponding to days 6-7 can be selected as the stable process data Aa for device a, i.e., Aa = A1 + A2 + A3. Similarly, a similar operation is performed on devices b, c, ... to obtain the stable process data Ab, Ac, ... for each of the m devices. Finally, the first process data is obtained based on the stable process data Ab, Ac, ... for each of the m devices.

[0069] It should be noted that, in some embodiments of this application, after obtaining the historical process data (i.e., stationary process data) included in each stable operating period (i.e., at least one sub-period) of each device, a difference operation can be performed on the historical process data included in each sub-period (all sub-periods of m devices) to obtain the timing information within each segment of stationary process data. Thus, each sub-period corresponds to a segment of stationary process data and a segment of timing information. All stationary process data and all timing information of all sub-periods constitute the aforementioned first process data.

[0070] It is important to note that if the historical process data included in a certain sub-period consists of p target data, then performing a difference operation on the historical process data included in that sub-period means performing a difference operation on those p specific target data, with the aim of establishing the temporal correlation between the various target data within that sub-period.

[0071] It should also be noted that, in the embodiments of this application, since stationary process data also includes multiple types, the difference operation is performed separately for each type of stationary process data in each sub-time period of each device.

[0072] 202. Calculate the slow feature transformation of the first process data to obtain the first transformation matrix of the slow feature.

[0073] Next, the slow feature transformation is performed on the first process data obtained above, resulting in the transformation matrix of the slow features, which can be called the first transformation matrix. In this step, the process of calculating the slow feature transformation is the process of performing SFA.

[0074] It is important to note that the first process data mentioned here is a fusion of stationary process data of n types (such as the three types in the example above), thereby achieving dimensionality reduction representation of multi-dimensional data. This first process data can be regarded as an initial matrix. The process of calculating the slow feature changes of this first process data is essentially: initial matrix * first transformation matrix = slow feature, and the first transformation matrix contains the most core key features of the initial matrix.

[0075] It should be noted that, in some embodiments of this application, in order to improve the sensitivity of the first transformation matrix, an exponential basis transformation can be performed on the first process data to obtain a second matrix, and then a slow feature transformation can be performed on the second matrix to obtain the first transformation matrix. The data contained in the slow features obtained in this way will have greater discriminative power.

[0076] 203. Deploy the first transformation matrix on the target device, and periodically update the first transformation matrix according to the second process data until the update termination condition is met. The second process data is obtained from the process data collected periodically on the target device, and the process data on the target device includes the n types of data.

[0077] The aforementioned first transformation matrix is ​​obtained based on historical process data of n types from m devices. That is, steps 201-202 above are training the first transformation matrix using all the historical process data from the m devices. This first transformation matrix can be used as the initial model when the device to be fault-detected (referred to as the target device) is powered on. In other words, the obtained first transformation matrix is ​​deployed on the target device. At this time, a higher alarm threshold can be set to prevent false alarms. Specifically, the operation can be as follows: collect a segment of initial process data from the target device's operation, calculate the slow feature corresponding to this initial process data, then calculate the upper bound of this slow feature (i.e., find the maximum value in a series of data), perform some calculations on this upper bound (such as multiplying by a factor) to obtain a strengthened upper bound, which can then be used as the alarm threshold.

[0078] It should be noted that the target device can be one of the aforementioned m devices, or it can be a device other than the m devices; this application does not limit this.

[0079] Then, the first transformation matrix can be periodically updated based on the second process data. This periodic update can be an incremental iterative update until the update termination condition is met. The process data on the target device includes the aforementioned n types of data, and the second process data is obtained based on the periodically collected process data on the target device. Specifically, the process data on the target device is periodically collected, and for each period's process data, similar processing to steps 201 and 202 above for the historical process data of the first historical time period (equivalent to a certain target period) (equivalent to the process data within that target period) is performed to obtain the second process data (equivalent to the first process data).

[0080] Taking period t1 as an example, assuming that process data T1 of period t1 of target device M is collected, the stable operating period (one or more) of period t1 is first determined. Then, the process data of the stable operating period is extracted from the process data T1. Differential operation is performed on the process data of each stable operating period. Differential operation is performed on each type of process data. Finally, the process data of each stable operating period of each type and the corresponding time sequence information are fused together to obtain the second process data.

[0081] As an example, this incremental update method can be seen in Figure 3. First, the initial first transformation matrix is ​​obtained by training the historical process data on old machine A and old machine B as the initial model. Then, the new machine C (i.e. the target device) uses a small amount of periodically collected process data to iteratively update the basic model to obtain a second transformation matrix specific to the new machine C as the specific model. Finally, the large amount of process data generated on the new machine C during operation, combined with the specific model, can obtain the final model.

[0082] It should be noted that in some embodiments of this application, the second process data includes process data for the current period and process data for historical periods, where the historical period refers to the period preceding the current period. One implementation for periodically updating the first transformation matrix based on the second process data is as follows: First, obtain the current transformation matrix increment based on the current period's process data. This current transformation matrix increment characterizes the influence of the current period's process data on the first transformation matrix updated in the previous period. The previous period refers to the period preceding the current period; for example, the period preceding period t2 is t1. Then, update the first transformation matrix updated in the previous period based on the current transformation matrix increment, thereby obtaining the updated first transformation matrix for the current period. This operation is performed for each current period until the update termination condition is met.

[0083] To better understand the above process, the following example will be used for illustration:

[0084] Assuming the target device is pump X, process data generated during the operation of pump X is periodically collected. In the first period t1, the collected process data is denoted as G1. First, the stable operating period of this first period t1 is determined. Then, process data from the stable operating period are extracted from process data G1 and subjected to differential operations (differential operations are required for each type of process data). Finally, the process data for each stable operating period of each type, along with the corresponding time sequence information, are obtained. These are then fused together to obtain the second process data R1 for period t1. Similarly, after periodically collecting the corresponding process data, the second process data R2 for period t2, the second process data R3 for period t3, ..., the second process data Ri for period ti can be obtained based on this operation.

[0085] In period t1, the first transformation matrix is ​​the initial first transformation matrix J0. Then, based on the second process data R1 in period t1, the current transformation matrix increment d can be obtained. J1 Then the first transformation matrix J1 = J0 + d is updated for the first time. J1 At this point, the first updated transformation matrix J1 is obtained based on the first process data and the second process data R1 with period t1.

[0086] Similarly, in period t2, the first transformation matrix is ​​the first transformation matrix J1 updated for the first time. Then, based on the second process data R2 in period t2, the current transformation matrix increment d can be obtained. J2 Then the first transformation matrix of the second update is J2 = J1 + d J2 At this point, the first transformation matrix J2 of the second update is obtained based on the first process data, the second process data R1 of period t1, and the second process data R2 of period t2.

[0087] And so on, the first transformation matrix J updated in the i-th iteration i =J i-1 +d Ji At this point, the first transformation matrix J updated for the i-th time... i It is obtained based on the first process data, the second process data R1 with period t1, the second process data R2 with period t2, ..., the second process data Ri with period ti.

[0088] When the first transformation matrix is ​​iteratively updated to the termination condition, the update stops, and the first transformation matrix obtained from the last update is the second transformation matrix.

[0089] It should be noted that, in some embodiments of this application, the conditions for terminating the update include, but are not limited to:

[0090] 1) The first transformation matrix obtained from the last update converges.

[0091] 2) The preset update duration is reached.

[0092] 3) Reach the preset update cycle.

[0093] In this step, the initial first transformation matrix is ​​incrementally updated using batch updates, incorporating real-time process data from the target device. This multi-batch, small-data update mode reduces storage and computational overhead while maintaining accuracy. Once the first transformation matrix reaches the termination condition, the alarm threshold can be adjusted to a value specific to this target device.

[0094] 204. When the value of the real-time target process data on the running target device is within the first range, the fault determination of the running target device is performed according to the second transformation matrix and the baseline value. The data type of the target process data is the target type, which is one of the n types. The second transformation matrix is ​​the first transformation matrix obtained in the last update. The baseline value is used to characterize the boundary line of the alarm.

[0095] Finally, fault determination can be performed on the operating target equipment based on the obtained second transformation matrix and baseline values. The data type of the target process data is the target type (one of the aforementioned n types). For example, when the target equipment is a pump, process data of the current RMS value type can be used as the target process data (i.e., operating condition indicator variable). The second transformation matrix is ​​the first transformation matrix obtained from the last update, and the baseline values ​​are used to characterize the alarm boundary. These baseline values ​​are obtained based on the second transformation matrix and the non-fault historical process data on the target equipment.

[0096] It is important to note that fault diagnosis requires confirming that the real-time target data generated by the operating target device falls within the first range. This is because the complex external environment can easily cause statistical quantities to briefly exceed limits, leading to false alarms. Therefore, a condition indicator variable is introduced to further refine the identification of system alarms. When the real-time target process data exceeds the baseline value after matrix transformation, the condition indicator variable is checked first. Only alarms that indicate stable operation (i.e., the value of the condition indicator variable falls within the first range representing stable operation) are considered valid alarms.

[0097] It should be noted that in some other embodiments of this application, if the value of the real-time target process data generated on the operating target device is not within the first range, the operation of fault determination of the target device is stopped. That is, if the operating condition indicator variable indicates a shutdown or non-stationarity (i.e., the value of the operating condition indicator variable falls within other value ranges that characterize a shutdown or non-stationarity), the alarm is canceled.

[0098] In summary, the improvements of the above embodiments of this application compared to existing solutions are mainly reflected in the following aspects (see Figure 4 for details):

[0099] (1) Feature enhancement

[0100] As shown in Figure 5, traditional methods determine equipment status by setting thresholds based on single-dimensional features. Typically, a very high threshold is required to cover changes in equipment production processes or large fluctuations in operating conditions, which can easily lead to missed detections.

[0101] The method in this application uses principal component analysis to obtain slow features to characterize the essential changes in the system. It incorporates multiple variables with different dimensions into a single model (i.e., the first transformation matrix) for unified processing, allowing for the consideration of equipment operating condition changes to a certain extent and further improving the model's robustness. More specifically, it can also extract the steady-state components of the equipment based on exponential basis transformation, further amplifying the weakly changing components in the system and extracting the slow-changing information of the system's essential characteristics, thus solving the problem of missed detections.

[0102] (2) Transformation matrix update (i.e. model update)

[0103] As shown in Figure 6, traditional model updates typically involve concatenating new and old data and then retraining to update the model. While this method preserves all valid information from the device, it also poses challenges to computation and storage. Especially at the edge, it is generally difficult to save all historical data.

[0104] The method in this application is based on the computing power of the edge control unit. It directly utilizes small batches of real-time process data and the base model (i.e., the second transformation matrix) at the edge to incrementally update the model in a "small amount, multiple times" manner, thereby completing the model retraining / incremental training. This significantly reduces the storage and computing power requirements at the edge while improving generalization ability.

[0105] (3) Double test

[0106] As shown in Figure 7, in the traditional scheme, a system alarm is generated when the statistical value exceeds the limit. Such alarms are usually difficult to exclude the influence of production conditions, operation mode switching, etc., and are therefore prone to false alarms.

[0107] The method in this application introduces a condition indicator variable to characterize the system's operating status. When an alarm occurs, an alarm that simultaneously satisfies the condition indicator variable being stationary is defined as a warning output, while an alarm that is non-stationary is not output. Therefore, fluctuations in statistical quantities are taken into account when considering the condition indicator variable, in order to eliminate false alarms.

[0108] Combining the above three aspects, a low-cost fault early warning and diagnosis scheme utilizing process control signals at the edge can be finally realized. Specifically, a schematic diagram of an overall process of the method in this application embodiment is shown in Figure 8. The above three aspects are respectively embodied in ①, ②, and ③ in Figure 8, which can be referred to in the above description and will not be repeated here.

[0109] Based on the above embodiments, in order to better implement the above solutions of the embodiments of this application, related equipment for implementing the above solutions is also provided below. Specifically, refer to Figure 9, which is a schematic diagram of a computer device provided in an embodiment of this application. The computer device 900 may specifically include: a determination module 901, a slow feature transformation module 902, an update module 903, and a fault determination module 904. The determination module 901 is used to obtain first process data based on n types of historical process data from m devices, where m≥1 and n≥2. The slow feature transformation module 902 is used to perform a slow feature transformation on the first process data to obtain a first transformation matrix of the slow features. The update module 903 is used to deploy the first transformation matrix on the target device and perform further processing on the first transformation matrix based on second process data. The process is periodically updated until the update termination condition is met. The second process data is obtained based on the process data collected periodically on the target device. The process data on the target device includes the n types of data. The fault determination module 904 is used to determine the fault of the target device in operation based on the second transformation matrix and the baseline value when the value of the real-time target process data on the target device in operation is within a first range. The data type of the target process data is the target type, which is one of the n types. The second transformation matrix is ​​the first transformation matrix obtained from the last update. The baseline value is used to characterize the boundary line of the alarm.

[0110] In one possible design, the second process data includes process data for the current cycle and process data for historical cycles, where the historical cycle is the cycle preceding the current cycle. The update module 903 is specifically used to: obtain the current transformation matrix increment based on the process data of the current cycle, where the current transformation matrix increment is used to characterize the influence of the process data of the current cycle on the first transformation matrix updated in the previous cycle, where the previous cycle is the cycle preceding the current cycle; and update the first transformation matrix updated in the previous cycle based on the current transformation matrix increment to obtain the first transformation matrix updated in the current cycle.

[0111] In one possible design, the baseline value is obtained based on the second transformation matrix and the non-fault history process data on the target device.

[0112] In one possible design, the determining module 901 is specifically used to: acquire n types of historical process data of m devices in a historical first time period; determine at least one sub-time period from the historical first time period, wherein the at least one sub-time period is a period in which the devices operate stably; and obtain first process data based on the historical process data in the at least one sub-time period.

[0113] In one possible design, the determining module 901 is further configured to: determine a first type from the n types; and determine at least one sub-period from the first historical period using historical process data belonging to the first type.

[0114] In one possible design, the determining module 901 is further configured to: determine at least one sub-period from the first historical period by performing an extended Dickie-Fowler test (ADF) on the historical process data belonging to the first type.

[0115] In one possible design, the determining module 901 is further configured to: perform differential operation on the historical process data in the at least one sub-period to obtain time series information, wherein the first process data includes the historical process data in the at least one sub-period and the time series information.

[0116] In one possible design, the slow feature transformation module 902 is specifically used to: perform an exponential basis transformation on the first process data to obtain a second matrix; and perform a slow feature transformation on the second matrix to obtain a first transformation matrix of the slow features.

[0117] In one possible design, the fault determination module 904 is also used to: stop performing fault determination on the target device if the value of the real-time target process data on the target device in operation is not within the first range.

[0118] In one possible design, the conditions for reaching the update termination include: the first transformation matrix obtained from the last update converges; or, a preset update duration is reached; or, a preset update cycle is reached.

[0119] It should be noted that the information interaction and execution process between the modules / units in the computer device 900 are based on the same concept as the method embodiments described above in this application. For details, please refer to the description in the method embodiments shown above in this application, which will not be repeated here.

[0120] Next, we will introduce another computer device provided in the embodiments of this application. Please refer to Figure 10. Figure 10 is a schematic diagram of the structure of a computer device provided in the embodiments of this application. The computer device 1000 may deploy the computer device 900 described in the corresponding embodiment of Figure 9, which is used to implement the functions of the computer device 900 in the corresponding embodiment of Figure 9. Specifically, the computer device 1000 is implemented by one or more servers. The computer device 1000 may vary greatly due to different configurations or performance. It may include one or more central processing units (CPUs) 1022 and memory 1032, and one or more storage media 1030 (e.g., one or more mass storage devices) for storing application programs 1042 or data 1044. The memory 1032 and storage media 1030 may be temporary storage or persistent storage. The program stored in the storage media 1030 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the computer device 1000. Furthermore, the central processing unit 1022 can be configured to communicate with the storage medium 1030 and execute a series of instruction operations in the storage medium 1030 on the computer device 1000.

[0121] Computer device 1000 may also include one or more power supplies 1026, one or more wired or wireless network interfaces 1050, one or more input / output interfaces 1058, and / or one or more operating systems 1041, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0122] In this embodiment, the central processing unit 1022 is used to execute the steps in the embodiment corresponding to FIG2. For example, the central processing unit 1022 can be used to: first, acquire historical process data of n types from m devices, and then obtain first process data based on the historical process data of n types, where m≥1 and n≥2. Next, perform a slow feature transformation on the obtained first process data to obtain a transformation matrix of the slow features, which can be called the first transformation matrix. Deploy the first transformation matrix on the target device, and then periodically update the first transformation matrix based on the second process data. This periodic update can be an incremental iterative update until the update termination condition is met. The process data on the target device includes the aforementioned n types of data, and the second process data is obtained based on the periodically collected process data on the target device. Finally, given that the real-time target process data on the running target device is within the first range, the running target device is fault-determined based on the obtained second transformation matrix and the baseline value. The data type of the target process data is the target type, which is one of the n types. The second transformation matrix is ​​the first transformation matrix obtained in the last update, and the baseline value is used to characterize the boundary line of the alarm.

[0123] It should be noted that the specific way in which the central processing unit 1022 executes the above steps is based on the same concept as the method embodiment corresponding to Figure 2 in this application, and the resulting technical effects are also the same as those in the above embodiments of this application. For details, please refer to the description in the method embodiment shown above in this application, which will not be repeated here.

[0124] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0125] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0126] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0127] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method for processing process data, characterized in that, include: The first process data is obtained from the historical process data of m devices of n types, where m≥1 and n≥2; Calculate the slow feature transformation on the first process data to obtain the first transformation matrix of the slow feature; The first transformation matrix is ​​deployed on the target device, and the first transformation matrix is ​​periodically updated according to the second process data until the update termination condition is met. The second process data is obtained from the process data collected periodically on the target device, and the process data on the target device includes the n types of data. When the real-time target process data on the running target device is within a first range, the running target device is fault-determined based on the second transformation matrix and the baseline value. The data type of the target process data is a target type, which is one of the n types. The second transformation matrix is ​​the first transformation matrix obtained from the last update. The baseline value is used to characterize the boundary line of the alarm.

2. The method according to claim 1, characterized in that, The second process data includes process data for the current period and process data for historical periods, wherein the historical periods are those preceding the current period. The step of periodically updating the first transformation matrix based on the second process data includes: The current transformation matrix increment is obtained based on the process data of the current period. The current transformation matrix increment is used to characterize the influence of the process data of the current period on the first transformation matrix updated in the previous period. The previous period is the period before the current period. The first transformation matrix updated in the previous period is updated based on the current transformation matrix increment to obtain the first transformation matrix updated in the current period.

3. The method according to any one of claims 1-2, characterized in that, The baseline value is obtained based on the second transformation matrix and the non-fault historical process data on the target device.

4. The method according to any one of claims 1-3, characterized in that, The process data obtained from n types of historical process data from m devices includes: Obtain historical process data of n types from m devices in the first historical time period; At least one sub-period is determined from the first historical period, wherein the at least one sub-period is a period in which the equipment operates stably; The first process data is obtained based on the historical process data in the at least one sub-period.

5. The method according to claim 4, characterized in that, Determining at least one sub-period from the first historical period includes: Determine the first type from the n types; At least one sub-period is determined from the first historical period using historical process data belonging to the first type.

6. The method according to claim 5, characterized in that, The step of determining at least one sub-period from the first historical period using historical process data belonging to the first type includes: At least one sub-period is determined from the first historical period by performing an Extended Dickie-Fowler test (ADF) on the historical process data belonging to the first type.

7. The method according to any one of claims 4-6, characterized in that, The step of obtaining the first process data based on historical process data in the at least one sub-time period includes: Differential operation is performed on the historical process data in the at least one sub-time period to obtain time series information, wherein the first process data includes the historical process data in the at least one sub-time period and the time series information.

8. The method according to any one of claims 1-7, characterized in that, The step of performing a slow feature transformation on the first process data to obtain the first transformation matrix of the slow features includes: The first process data is subjected to an exponential basis transformation to obtain the second matrix; The slow feature transformation is obtained by applying the slow feature transformation to the second matrix, resulting in the first transformation matrix of the slow feature.

9. The method according to any one of claims 1-8, characterized in that, The method further includes: If the value of the real-time target process data on the target device in operation is not within the first range, the fault determination of the target device shall be stopped.

10. The method according to any one of claims 1-9, characterized in that, The conditions for terminating the update include: The first transformation matrix obtained from the last update converges; or, The preset update duration has been reached; or, The preset update cycle has been reached.

11. A computer device, characterized in that, include: The determination module is used to obtain the first process data based on n types of historical process data from m devices, where m≥1 and n≥2. The slow feature transformation module is used to calculate the slow feature transformation on the first process data to obtain the first transformation matrix of the slow feature. An update module is used to deploy the first transformation matrix on the target device and periodically update the first transformation matrix according to the second process data until the update termination condition is met. The second process data is obtained from the process data collected periodically on the target device, and the process data on the target device includes the n types of data. The fault determination module is used to determine the fault of the target device in operation based on a second transformation matrix and a baseline value when the real-time target process data on the target device in operation is within a first range. The data type of the target process data is a target type, which is one of the n types. The second transformation matrix is ​​the first transformation matrix obtained from the last update. The baseline value is used to characterize the boundary line of the alarm.

12. The device according to claim 11, characterized in that, The second process data includes process data for the current cycle and process data for historical cycles, wherein the historical cycles are those preceding the current cycle. The update module is specifically used for: The current transformation matrix increment is obtained based on the process data of the current period. The current transformation matrix increment is used to characterize the influence of the process data of the current period on the first transformation matrix updated in the previous period. The previous period is the period before the current period. The first transformation matrix updated in the previous period is updated based on the current transformation matrix increment to obtain the first transformation matrix updated in the current period.

13. The device according to any one of claims 11-12, characterized in that, The baseline value is obtained based on the second transformation matrix and the non-fault historical process data on the target device.

14. The device according to any one of claims 11-13, characterized in that, The determining module is specifically used for: Obtain historical process data of n types from m devices in the first historical time period; At least one sub-period is determined from the first historical period, wherein the at least one sub-period is a period in which the equipment operates stably; The first process data is obtained based on the historical process data in the at least one sub-period.

15. The device according to claim 14, characterized in that, The determining module is further configured to: Determine the first type from the n types; At least one sub-period is determined from the first historical period using historical process data belonging to the first type.

16. The device according to claim 15, characterized in that, The determining module is further configured to: At least one sub-period is determined from the first historical period by performing an Extended Dickie-Fowler test (ADF) on the historical process data belonging to the first type.

17. The device according to any one of claims 14-16, characterized in that, The determining module is further configured to: Differential operation is performed on the historical process data in the at least one sub-time period to obtain time series information, wherein the first process data includes the historical process data in the at least one sub-time period and the time series information.

18. The device according to any one of claims 11-17, characterized in that, The slow feature transformation module is specifically used for: The first process data is subjected to an exponential basis transformation to obtain the second matrix; The slow feature transformation is obtained by applying the slow feature transformation to the second matrix, resulting in the first transformation matrix of the slow feature.

19. The device according to any one of claims 11-18, characterized in that, The fault determination module is also used for: If the value of the real-time target process data on the target device in operation is not within the first range, the fault determination of the target device shall be stopped.

20. The device according to any one of claims 11-19, characterized in that, The conditions for terminating the update include: The first transformation matrix obtained from the last update converges; or, The preset update duration has been reached; or, The preset update cycle has been reached.

21. A computer device comprising a processor and a memory, the processor being coupled to the memory, characterized in that, The memory is used to store programs; The processor is configured to execute a program in the memory, causing the computer device to perform the method as described in any one of claims 1-10.

22. A computer storage medium, characterized in that, The device stores computer-readable instructions, which, when executed by a processor, implement the method as described in any one of claims 1-10.

23. A computer program product, characterized in that, The computer program product includes computer-readable instructions that, when executed by a processor, implement the method as described in any one of claims 1-10.

24. A chip, the chip comprising a processor and a data interface, characterized in that, The processor reads instructions stored in the memory through the data interface and executes the method as described in any one of claims 1-10.

Images