Seamless steel tube production line real-time monitoring system

Abstract

The invention discloses a seamless steel tube production line real-time monitoring system, which relates to the technical field of real-time monitoring, and comprises the following steps: collecting multi-source heterogeneous state data of a multi-channel sensor, synchronously caching the multi-source heterogeneous state data in an edge monitoring node, and generating a standardized state data sequence; key energy feature dimensions are extracted according to the standardized state data sequence, energy mapping indexes are constructed, and a dynamic soft threshold boundary of monitoring parameters is generated based on historical production samples; and comparing the dynamic soft threshold boundary with a real-time observation value at a time point corresponding to the dynamic soft threshold boundary in the standardized state data sequence, extracting a preliminary anomaly mark, and screening out misjudgment anomalies by using a pseudo anomaly shielding mechanism to generate a credible anomaly layer. According to the method, a prediction driving method is adopted, and an optimized production parameter adjustment instruction is generated according to a dynamic risk grade grading interface and a standardized state data sequence, so that parameter adjustment strategy reasoning taking a risk evolution trend as a core drive is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of real-time monitoring, in particular to a seamless steel tube production line real-time monitoring system. BACKGROUND

[0002] In the field of seamless steel tube production, in order to protect product quality and production efficiency, multi-channel sensors are generally deployed in each production link to obtain state variables such as temperature, pressure, vibration and current, and industrial control networks or field buses are used for data aggregation, and then real-time monitoring and fault warning are performed by upper monitoring devices. In the traditional method, the collected data is analyzed and judged by setting fixed thresholds or empirical models, so as to realize online monitoring and early warning of key production parameters. In addition, some monitoring systems introduce data visualization interfaces to convert multi-source state information into chart forms for display, so as to improve the understanding and response ability of operators to the current working condition. These methods have been widely applied in industrial sites and constitute the mainstream technical path of current seamless steel tube production line digital monitoring.

[0003] However, the conventional method still faces challenges in dynamic adaptability and abnormality recognition credibility. On the one hand, the conventional method usually relies on static threshold setting, which is difficult to adapt to the dynamic changes brought by the fluctuation of working condition parameters over time, and cannot effectively perceive the risk change trend in the nonlinear evolution process; on the other hand, the abnormality detection mechanism is mostly based on single-point discrimination logic, which is easily disturbed by data jitter or transient abnormal values, resulting in frequent misjudgment, and thus affecting the credibility of the abnormality result and the accuracy of subsequent parameter adjustment. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a seamless steel tube production line real-time monitoring system to solve the problems of insufficient risk perception ability and low abnormality recognition credibility under dynamic changes of working condition parameters.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a seamless steel tube production line real-time monitoring system, which comprises,

[0008] a data acquisition module, which acquires multi-source heterogeneous state data of multi-channel sensors and synchronously buffers in an edge monitoring node to generate a standardized state data sequence;

[0009] a feature modeling module, which extracts key energy feature dimensions according to the standardized state data sequence, constructs an energy mapping index, and generates a dynamic soft threshold boundary of the monitoring parameter based on historical production samples;

[0010] An anomaly screening module compares the dynamic soft threshold boundary with the real-time observation value at the time point corresponding to the dynamic soft threshold boundary in the standardized state data sequence, extracts a preliminary anomaly marker, screens out false anomaly using a pseudo-anomaly shielding mechanism, and generates a credible anomaly layer;

[0011] A layer generation module dynamically generates a production index rendering layer interface in combination with the standardized state data sequence and the preliminary anomaly marker, and dynamically generates a risk perception interface based on the fusion calculation of the standardized state data sequence and the preliminary anomaly marker;

[0012] A risk grading module fuses the credible anomaly layer into the risk perception interface, and generates a dynamic risk level grading interface in combination with the production index rendering layer interface;

[0013] A parameter optimization module generates an optimized production parameter adjustment instruction based on the dynamic risk level grading interface and the standardized state data sequence.

[0014] As a preferred scheme of the seamless steel pipe production line real-time monitoring system, the multi-source heterogeneous state data of the multi-channel sensor is collected and synchronously buffered in the edge monitoring node to generate a standardized state data sequence, and the steps are as follows,

[0015] The multi-channel sensor of the seamless steel pipe production line is deployed to collect multi-source heterogeneous state data through a wireless protocol access edge collection unit, and a structured data tuple sequence is constructed by uniformly decoding through an analysis template;

[0016] Based on the structured data tuple sequence, the collection time is aligned by using a precise clock synchronization protocol, and a sliding window parameter is set to construct an original state window sequence.

[0017] Each channel variable in the original state window sequence is subjected to Z-score standardization processing, and is converted into a standardized state data sequence according to historical statistical characteristics.

[0018] As a preferred scheme of the seamless steel pipe production line real-time monitoring system, the precise clock synchronization protocol is periodically exchanged between the master clock and the slave clock through the timestamp information to obtain the transmission delay and the clock offset.

[0019] As a preferred scheme of the seamless steel pipe production line real-time monitoring system, the key energy feature dimension is extracted according to the standardized state data sequence, an energy mapping index is constructed, and a dynamic soft threshold boundary of the monitoring parameter is generated based on historical production samples, and the steps are as follows,

[0020] Based on the standardized state data sequence, the energy feature index is generated by using a sliding window processing, and the energy feature set in a unified structure is obtained by integration.

[0021] The redundant constant dimensions in the energy feature set are removed, and mutual information and MIC analysis are used to combine into an energy feature dimension set;

[0022] Each dimension energy feature in the energy feature dimension set is combined by a weighted fusion strategy to form an energy mapping index sequence.

[0023] Based on the cyclic fluctuation sequence in the energy mapping index sequence, an autoregressive integrated moving average model is constructed by combining the energy mapping index values in the historical production samples, then a dynamic threshold compensation is generated under a set confidence level, and finally the energy mapping index sequence and the dynamic threshold compensation value are superimposed to generate a dynamic soft threshold boundary.

[0024] As a preferred scheme of the seamless steel pipe production line real-time monitoring system, wherein: the dynamic soft threshold boundary is compared with the real-time observation value of the corresponding time point of the dynamic soft threshold boundary in the standardized state data sequence, the preliminary abnormality mark is extracted, the false abnormality is screened out by using the pseudo abnormality shielding mechanism, and the credible abnormality layer is generated, and the steps are as follows,

[0025] The real-time observation value of each channel in the standardized state data sequence is compared with the upper and lower limits of the dynamic soft threshold boundary, and the preliminary abnormality mark set is generated by summarizing;

[0026] The preliminary abnormality mark set is screened by pseudo abnormality judgment, the false abnormality caused by non-real factors is removed, and the credible abnormality mark set is generated after screening and correction;

[0027] According to the credible abnormality mark set, the time point and channel number of the marked abnormality in the standardized state data sequence are located, the credible abnormality layer construction field set in each group of time points and channel numbers is extracted, and the credible abnormality layer is organized and constructed in a structured data form.

[0028] As a preferred scheme of the seamless steel pipe production line real-time monitoring system, wherein: the pseudo abnormality judgment process refers to identifying and removing the false abnormality marks caused by non-real state changes in the preliminary abnormality mark set.

[0029] As a preferred scheme of the seamless steel pipe production line real-time monitoring system, wherein: the field structure of the credible abnormality layer construction field set includes channel number, time stamp, standardized state variable value, energy mapping index value, upper and lower limits of dynamic soft threshold boundary, and energy feature dimension contribution information corresponding to the state variable.

[0030] As a preferred scheme of the seamless steel tube production line real-time monitoring system, wherein: the standardized state data sequence and the preliminary anomaly mark are combined, a production index rendering layer interface is dynamically generated, and a risk perception interface is dynamically generated based on the fusion calculation of the standardized state data sequence and the preliminary anomaly mark, and the steps are as follows,

[0031] The state values of all channels and time points in the standardized state data sequence and the preliminary anomaly mark are mapped to form an index rendering matrix.

[0032] Based on the index rendering matrix, the dynamic drawing is performed by using the layer rendering method to generate the production index rendering layer interface.

[0033] The risk score sequence is calculated by combining the preliminary anomaly mark statistical anomaly distribution characteristics and the weighted fusion method, and the risk score sequence is fused with the production index rendering layer interface to generate the risk perception interface.

[0034] As a preferred scheme of the seamless steel tube production line real-time monitoring system, wherein: the trusted anomaly layer is fused into the risk perception interface, and a dynamic risk level classification interface is generated in combination with the production index rendering layer interface, and the steps are as follows,

[0035] The abnormal positioning information set is extracted from the trusted anomaly layer and integrated to build an abnormal channel time distribution table;

[0036] Based on the abnormal channel time distribution table, the risk score data in the risk perception interface, and the abnormal positioning information in the trusted anomaly layer, an abnormal propagation path graph is constructed, and then the risk level threshold interval is dynamically divided according to the correlation strength of the abnormal propagation path graph and the risk score data;

[0037] The risk level threshold interval is fused with the production index rendering layer interface by using the risk level graphical coding method to obtain a rendering layer interface with superimposed risk level labels;

[0038] Based on the rendering layer interface with superimposed risk level labels, the rendering layer interface is dynamically updated by using the time series driving method to form the dynamic risk level classification interface.

[0039] As a preferred scheme of the seamless steel tube production line real-time monitoring system, wherein: based on the dynamic risk level classification interface and the standardized state data sequence, an optimized production parameter adjustment instruction is generated, and the steps are as follows,

[0040] A prediction driving model is constructed based on the dynamic risk level classification interface and the standardized state data sequence, and a future state-risk prediction sequence is generated by reasoning;

[0041] An optimal adjustment strategy of a future state-risk prediction sequence and a working condition parameter constraint set is solved to construct a working condition parameter adjustment solution space;

[0042] According to a preset optimization objective function, the working condition parameter adjustment solution space is optimized to obtain an optimized production parameter adjustment instruction.

[0043] The present application has the beneficial effects that: by adopting the prediction driving method, the steps of generating the optimized production parameter adjustment instruction according to the dynamic risk level grading interface and the standardized state data sequence, the parameter adjustment strategy reasoning driven by the risk evolution trend is realized, the adjustment instruction generation mechanism combining the future risk situation change and the current process state is constructed, thus the forward-looking intervention and risk control of the seamless steel pipe production process are realized, and the beneficial effects of guaranteeing the production continuity and reducing the process failure rate are finally achieved. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced as follows, and obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0045] Fig. 1 Flow chart of the seamless steel pipe production line real-time monitoring system.

[0046] Fig. 2 Feature modeling module flow chart of the seamless steel pipe production line real-time monitoring system.

[0047] Fig. 3 Abnormal screening module flow chart of the seamless steel pipe production line real-time monitoring system.

[0048] Fig. 4 Parameter optimization module flow chart of the seamless steel pipe production line real-time monitoring system. DETAILED DESCRIPTION

[0049] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0050] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited by the specific embodiments disclosed below.

[0051] Second, the "one embodiment" or "an embodiment" referred to herein can include a particular feature, structure, or characteristic. The various embodiments appearing at different places in this specification can not all refer to the same embodiment or to the same particular feature, structure, or characteristic. Moreover, it is appreciated that features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0052] Referring to Figs. 1-4 For one embodiment of the present application, the embodiment provides a seamless steel tube production line real-time monitoring system, comprising the following steps:

[0053] The data acquisition module acquires multi-source heterogeneous state data of the multi-channel sensor and synchronously buffers in the edge monitoring node to generate a standardized state data sequence, comprising the following steps,

[0054] The multi-channel sensor is deployed in the seamless steel tube production line, and the collected state quantity data is transmitted to the edge acquisition unit interface through a wireless communication protocol channel. In the data acquisition process, each data channel is identified as a physical access path corresponding to a sensor, which is used to receive real-time observation data collected by the sensor, and the whole constitutes a multi-source heterogeneous state data set. Subsequently, a preset state data parsing template library is called, and according to the transmission format and structure specification corresponding to each type of multi-source heterogeneous state data, field-level data parsing processing and unified format standard mapping operations are performed. The data parsed by the state data parsing template library is uniformly converted into a standardized state data tuple set with a fixed field structure, and all standardized state data tuples are arranged in accordance with the sampling time axis sequence to finally build a structured state data tuple sequence.

[0055] The multi-source heterogeneous state data includes but is not limited to: process data (such as temperature, pressure, speed, and tensile force, etc. Key process engineering physical quantities), equipment running data (such as motor current, voltage, bearing vibration, and hydraulic pressure, etc. Device health status), environmental background data (such as indoor and outdoor air temperature, humidity, and dust concentration, etc. Peripheral environmental condition data), position and motion state data (such as real-time position, speed, and displacement of the steel tube in the conveying line, etc. Spatial dynamic data), image and visual data feature values (such as defect size, edge intensity, and contrast index, etc. Structure characteristic quantity extracted from the surface quality image obtained by the industrial camera), and control input state data (such as gate opening, actuator state, and PLC instruction feedback value, etc. Data of configuration state).

[0056] The preset state data analysis template library is a set of field-level data analysis rules that are constructed in advance based on the data output format and semantic structure of various types of sensors in the seamless steel pipe production line. Each set of state data analysis templates includes a description of the name, data type, unit conversion, and standardized mapping method of the original data field. The state data analysis templates are indexed by device type or communication protocol, and unified analysis and standardized conversion of state data of different sources and structures can be achieved through template library calling.

[0057] Based on the structured data tuple sequence, the collection time stamps carried by each channel data in the structured data tuple sequence are uniformly aligned using a precise clock synchronization protocol. During the alignment process, the time stamps in each structured state data tuple are time-shifted and corrected according to the global synchronization time reference in the precise clock synchronization protocol, eliminating the time errors caused by network delay or local clock drift between channels. Subsequently, the structured state data tuple sequence after time alignment is used as the basis for constructing a sliding window. By setting the sliding window parameters, continuous original state segments are extracted from the structured state data tuple sequence in chronological order, and an original state window sequence is generated in turn.

[0058] The precise clock synchronization protocol is a high-precision time synchronization protocol designed for industrial field and distributed data collection environment, which can align the clocks between devices in a local area network. The precise clock synchronization protocol obtains the transmission delay and clock offset by periodically exchanging time stamp information between the master clock and the slave clock.

[0059] In the original state window sequence, the corresponding process and physical state variables (such as temperature, pressure, and vibration, etc.) in each data channel are respectively subjected to Z-score standardization processing. Subsequently, the standardized data channel variables are organized and arranged in chronological order according to the position of the sliding window on the time axis, forming a standardized state data sequence.

[0060] The Z-score formula is as follows,

[0061]

[0062] where i represents the number of sampling time points, j represents the number of data channels, z i,j represents the standardized data of the jth channel at the ith time point in the standardized state data sequence, x i,j represents the original state quantity observation value of the jth channel at the ith time point in the structured data tuple sequence, μ j represents the historical mean of the jth channel, and σ j represents the historical standard deviation of the jth channel.

[0063] The feature modeling module extracts key energy feature dimensions from the standardized state data sequence, constructs an energy mapping index, and generates a dynamic soft threshold boundary for the monitoring parameters based on historical production samples, including the following steps,

[0064] Based on the standardized state data sequence, each state variable in the standardized state data sequence is processed by a preset sliding window parameter for channel-by-channel window cutting. For the standardized state value sequence in each sliding window, the energy feature index is obtained by square accumulation summation and mean normalization. Finally, the energy feature indexes corresponding to each sliding window position of all state variables are aligned and integrated according to the channel number and time position to generate an energy feature set with a unified structure.

[0065] It should be explained that the preset sliding window parameter includes the sliding window width and the sliding step, which is set according to the time resolution of the standardized state data sequence and the dynamic characteristic time length of the target detection event.

[0066] The energy feature set is traversed and calculated according to the global variance to generate a volatility index, and a fixed variance threshold is set. Energy feature dimensions below the fixed variance threshold are removed to remove redundant constant dimensions with minimal numerical fluctuations in all sampling time periods. Then, the energy feature set after removing the constant dimensions is subjected to information correlation analysis operation. The mutual information method is used to calculate the information dependence between each pair of energy feature dimensions and the target variable, and the maximum information coefficient (MIC) method is used to describe the strength of nonlinear correlation. Finally, through the dual analysis results of mutual information and maximum information coefficient, energy feature dimensions with strong correlation and information complementarity are selected and combined to form an energy feature dimension set with a unified structure.

[0067] The traversal calculation formula is as follows,

[0068]

[0069] Wherein, r represents the energy feature dimension number, N represents the total number of time points in the standardized state data sequence, k represents the time point number traversed in the current summation operation, m represents the time point number in the mean calculation, represents the value of the rth energy feature dimension at the kth time point, represents the value of the rth energy feature dimension at the mth time point, V r represents the volatility index of the rth energy feature dimension at all time points.

[0070] It needs to be explained that the fixed variance threshold is used to identify and eliminate the redundant feature dimensions in the energy feature set that fluctuate very little and have no obvious change in the entire sampling period. The basis for setting the fixed variance threshold is to calculate the global variance of all energy feature dimensions, analyze the distribution of energy feature dimensions in the sample data, and eliminate according to the preset quantile or empirical threshold.

[0071] For example: taking a typical industrial sample as an example, suppose that 80 energy feature dimensions are extracted, and after statistical analysis of the variances of all feature dimensions, it is found that the variance distribution of most effective feature dimensions is concentrated in the interval of 0.005 to 0.05, and there are some feature dimensions with variances below 0.001, with very small fluctuation amplitude. At this time, the fixed variance threshold can be set to 0.001, and all feature dimensions with variances below the threshold are eliminated.

[0072] Based on the energy feature dimension set, the relative contribution degree of each energy feature dimension to the overall state change is used to construct a weight coefficient vector. Then, according to the arrangement order of the feature dimensions in the energy feature dimension set, the weighted sum and fusion of each energy feature value in the energy feature dimension set and the corresponding weight coefficient are performed, and finally the energy mapping index sequence is generated in time sequence.

[0073] Among them, the relative contribution degree of the overall state change contribution refers to the weight proportion of each dimension in the overall state change among multiple energy feature dimensions. Usually, the relative contribution degree is obtained by evaluating feature contribution based on variance explanation rate, mutual information correlation coefficient, and machine learning model (such as linear regression coefficient, tree model, etc.).

[0074] Based on the energy mapping index sequence, a time series decomposition method is used to separate long-term trend sequence, seasonal variation sequence, cyclic fluctuation sequence and irregular fluctuation sequence, and combined with historical generated samples, an ARIMA model is constructed based on the cyclic fluctuation sequence, and then a dynamic threshold compensation amount is generated at a confidence level. The long-term trend sequence and the dynamic threshold compensation amount are linearly superimposed to generate a dynamic soft threshold boundary.

[0075] Among them, the time series decomposition method is a statistical technique that decomposes original time series data into multiple interpretable components. The core goal is to separate the deterministic pattern and random fluctuation in the energy mapping index sequence, reveal the internal structure and evolution law of the energy mapping index sequence, and decompose the energy mapping index sequence into four key components: long-term trend sequence, seasonal variation sequence, cyclic fluctuation sequence and irregular fluctuation sequence.

[0076] It needs to be explained that the ARIMA model is the core input of the cyclic fluctuation sequence extracted from the time series decomposition. Through the autoregressive term to capture the continuous influence of historical data, the difference term to eliminate non-stationary fluctuations, and the moving average term to correct the prediction error, the ARIMA model framework is formed. And in the ARIMA model training process, the stationary check and difference processing are carried out on the cyclic fluctuation sequence, and the order of the ARIMA model is determined according to the autocorrelation and partial autocorrelation graph. Then the maximum likelihood estimation is used to solve the optimal coefficient, and the training of the ARIMA model is completed.

[0077] It needs to be explained that the historical generation sample refers to the standardized state data set collected in the historical normal operation period. The core role is to establish a benchmark reference model for normal operation state, which includes the following data: energy mapping index sequence, time series decomposition component, working condition parameter metadata and quality association label.

[0078] The abnormality screening module compares the dynamic soft threshold boundary with the real-time observation value of the corresponding time point in the standardized state data sequence, extracts the preliminary abnormality mark, and uses the pseudo abnormality shielding mechanism to screen out the misjudgment abnormality, generates a credible abnormality layer, including the following steps,

[0079] For the real-time observation value of each channel in the standardized state data sequence, the upper and lower limits of the dynamic soft threshold boundary at the corresponding time are compared one by one to determine whether there is a situation of exceeding the interval range. Specifically, if the channel observation value at a certain time is less than the lower limit of the dynamic soft threshold boundary at the corresponding time, or greater than the upper limit of the dynamic soft threshold boundary at the corresponding time, it is recorded as the abnormal mark of the corresponding sensor signal channel at that time. All the above judgment processes are executed in turn in all channels and all time series ranges in the standardized state data sequence, and all the marked abnormal channels are summarized to finally form a preliminary abnormality mark set.

[0080] Based on each abnormality mark in the preliminary abnormality mark set, combined with the change trend between the standardized state data of each abnormality mark corresponding channel in the adjacent time period, the evolution trajectory of the energy mapping index sequence over time, and the normal fluctuation characteristics in the historical production sample, the pseudo abnormality judgment process is executed, the misjudgment situation caused by short-term fluctuation, measurement error or frequent switching is identified, the abnormal mark judged as pseudo abnormality is excluded, the abnormal mark not judged as pseudo abnormality is retained, and according to the time sequence continuity and adjacent channel cooperative fluctuation information in the judgment process, part of the abnormal boundary position is modified, and finally a credible abnormality mark set is generated.

[0081] Wherein, the pseudo-abnormality determination process is to identify and remove the misjudgment abnormality markers caused by non-true state changes in the preliminary abnormality marker set, for example, at the t=35 time point, the channel energy mapping index exceeds the dynamic soft threshold, but at t=34 and t=36 time points, it is still within the dynamic soft threshold normal range, and the local fluctuation is minimal, such mutation points are determined as pseudo-abnormality markers, which are removed and no longer used as abnormality input for downstream analysis.

[0082] Wherein, the time sequence continuity is the continuity and trend consistency of the abnormal state of the energy mapping index sequence on the time axis at a plurality of continuous time points; the adjacent channel coordinated fluctuation information refers to whether there is a synchronous or coordinated direction energy mapping index abnormal fluctuation phenomenon in the adjacent time period in the physically or functionally adjacent observation channels.

[0083] According to the trusted abnormality marker set, each marker as trusted abnormality observation data is searched in the standardized state data sequence, and the time point and channel number information corresponding to the trusted abnormality observation data are determined, and the time channel positioning coordinates are formed by the abnormal event index mapping method. Subsequently, for each set of time channel positioning coordinates, the corresponding state variable field information, energy mapping index value and context information of the corresponding time point are extracted from the standardized state data sequence, the trusted abnormality layer construction field set is constructed, and finally all the trusted abnormality layer construction field sets are arranged in time sequence, and finally organized as a structured data set of the trusted abnormality layer.

[0084] It needs to be explained that the abnormal event index mapping method is based on the abnormal state variable marker recorded in the trusted abnormality marker set, which performs a two-dimensional indexing operation on the standardized state data sequence, that is, simultaneously locates according to time index and channel number index, accurately identifies the two-dimensional position coordinates of abnormal observation in the state data sequence, thereby generating a data positioning coordinate set for subsequent field extraction and layer construction.

[0085] Wherein, the trusted abnormality layer construction field set is organized according to a unified field structure, and the field structure includes: channel number, time stamp, standardized state variable value, energy mapping index value, dynamic soft threshold upper and lower limit, and state variable corresponding energy feature dimension contribution information.

[0086] The layer generation module dynamically generates a production index rendering layer interface in combination with the standardized state data sequence and the preliminary abnormality marker, and dynamically generates a risk perception interface based on the fusion calculation of the standardized state data sequence and the preliminary abnormality marker, including the following steps,

[0087] Based on the standardized state data sequence and the preliminary abnormal marker set, all marked time point and channel number coordinate positions in the preliminary abnormal marker set and the standardized state values corresponding to the positions of each time point and channel number in the standardized state data sequence are extracted, and an index state value set is formed in a position index mapping extraction manner. Subsequently, the index state value set is matrix-organized in a time point as row and channel number as column arrangement manner, and finally an index rendering matrix is constructed in the form of an arrangement structure of time point as row and channel number as column with the index state value set as content.

[0088] The position index mapping extraction manner refers to using the time point and channel number recorded in the preliminary abnormal marker set as two-dimensional indexes to locate and extract the state values of the corresponding positions in the standardized state data sequence one by one, and then construct a data set organized according to the index correspondence.

[0089] According to the structure order of time point as horizontal axis and channel number as vertical axis, each index state value in the index rendering matrix (for example, the index state value size is represented by color depth, brightness intensity or texture change) is subjected to graphic coding processing, then all the graphically coded index state values are dynamically drawn in the form of a single frame layer according to the layer rendering manner, and a layer sequence is sequentially generated by superimposition according to the time dimension order, and finally a continuously changing production index rendering layer interface is constructed through the time series layer superimposition rendering mechanism.

[0090] It should be explained that the time series layer superimposition rendering mechanism refers to sequentially superimposing a plurality of layer data frames generated according to the time sequence order (for example: each frame is an index state layer of a time point) according to the time order, using the method of continuous frame refreshing and rendering, to construct a static layer into a layer interface with dynamic evolution effect.

[0091] It should be explained that the layer rendering manner is a professional drawing method for visualizing and presenting multi-dimensional and multi-source index state values in a layered and superimposed form in a two-dimensional or three-dimensional visualization interface. For example, in a heat map manner, the standardized state values are mapped to color values (such as from color-1 to red+1), different time points form different layer frames, and a dynamic layer display interface is formed through time axis playing.

[0092] Based on the preliminary abnormal marker set, the number of abnormal channels appearing at each time point is counted to obtain the abnormal event distribution characteristics of each time point. Subsequently, according to the proportion of the number of abnormal channels, the channel distribution weight and the abnormal intensity factor and other weighted factors, a weighted fusion method is used to calculate the risk score value of each time point to form a risk score sequence, and the risk score sequence is time-synchronously fused with the production index rendering layer interface, and rendered and mapped through frame-level layer superimposition, and finally a risk perception interface is dynamically generated.

[0093] The calculation formula of the weighted fusion method is as follows,

[0094]

[0095] Wherein, τ represents the time point variable of the current calculation risk score, l represents the abnormal channel number variable, represents the risk score value at time point τ, M τ represents the total number of abnormal channels involved in time point τ, β l represents the importance weight of channel number l, η τ,l represents the abnormal intensity value of time point τ on abnormal channel l, γ l represents the abnormal level influence coefficient of abnormal channel l.

[0096] It needs to be explained that the frame-level layer superposition mode refers to a kind of visual drawing mode that different data layers with the same time stamp are superimposed on the real scene according to the preset rendering order at the same time point. For example, in the real-time monitoring interface of seamless steel pipe production line, the temperature layer, equipment vibration layer and risk level layer can be superimposed and displayed according to the rendering order of "temperature first, vibration in the middle and risk on top" at the same time, so as to realize the synchronous visualization of multi-source state information.

[0097] The risk grading module fuses the trusted abnormal layer into the risk perception interface, generates a dynamic risk level grading interface in combination with the production index rendering layer interface, including the following steps,

[0098] In the trusted abnormal layer, the abnormal channel number and the corresponding abnormal time point information marked therein are extracted one by one to form an abnormal positioning information set containing channel number and time point. Then, the abnormal positioning information set is grouped according to the channel number, and the time points in each group are sequentially arranged according to the time point sequence. Finally, an abnormal channel time distribution table is constructed, taking the channel number as the main index and the time point sequence as the content.

[0099] It needs to be explained that the extraction one by one refers to the extraction operation of each abnormal positioning item composed of time point and abnormal channel number in the trusted abnormal layer.

[0100] After the construction of the abnormal channel time distribution table is completed, the abnormal propagation path graph is constructed based on the abnormal positioning information in the trusted abnormal layer, and the correlation strength of the abnormal propagation path graph is quantified. Then, the correlation strength and the risk score data are dynamically calculated by the weighted fusion strategy to obtain the risk weight coefficient, and the dynamic threshold interval is divided in real time according to the risk weight coefficient. The risk score value of each time point is mapped to the risk level category in the corresponding dynamic interval.

[0101] It needs to be explained that the abnormal propagation path diagram is a directed topological network for describing the propagation path and influence intensity of abnormal events of the production line among multiple process nodes, and the core function is to locate the abnormal source through visual topological structure, and the nodes of the abnormal propagation path diagram are key monitoring points of the production process (for example: rolling mill temperature, cooling water pressure, bearing vibration and other process parameter collection points).

[0102] In the risk level time sequence, the risk level identifier corresponding to each time point is processed by graphical coding, and then at each time point, the graphically coded risk level identifier is superimposed on the corresponding time point position in the production index rendering layer interface according to the time axis order, and the layer superposition order is kept consistent, and finally the rendering layer interface with the superimposed risk level label is obtained.

[0103] It needs to be explained that the graphical coding processing is to assign a unique graphical identifier to different risk level categories under the preset graphical coding rule, for example, using color, shape, size and other dimensions for coding, and the coding method is based on the preset graphical label rule to realize clear distinction and rapid identification of risk levels in the visual interface, for example, the first level risk level identifier is coded as a red dot, the second level risk level identifier is coded as an orange triangle, etc.

[0104] Among them, the preset graphical coding rule refers to the mapping relationship between a set of graphical features and risk levels for different risk level categories in the visual rendering process, which is set according to the importance of risk level and visual differentiation needs to ensure that different levels have obvious differences in vision. For example, the preset rule can set the first level risk level to correspond to red circle, the second level risk level to correspond to orange triangle, and the third level risk level to correspond to yellow square, with color from dark to light and shape from simple to complex to enhance the distinguishability and expression clarity of risk levels in the layer fusion process.

[0105] In the rendering layer interface with the superimposed risk level label, based on the time axis order, the rendering layer frame corresponding to each time point in the rendering layer interface with the superimposed risk level label is retrieved in turn, and then each frame layer content is loaded and displayed in turn through the visual interface to present the risk level change of each time point in time sequence, thereby generating a dynamic risk level classification interface covering a continuous period.

[0106] It needs to be explained that the rendering layer frame refers to the image content generated by layer fusion of the production index rendering layer and the risk level graphical coding identifier through frame-level layer superposition at a certain time point.

[0107] The frame-level layer superposition mode refers to, at the same time point, superimposing and drawing, layer by layer, data layers with different sources but consistent timestamps according to a preset layer rendering order to form a composite image containing multiple visualization information dimensions.

[0108] The parameter optimization module generates an optimized production parameter adjustment instruction based on the dynamic risk level classification interface and the standardized state data sequence, including the following steps,

[0109] The time series supervised learning method is adopted to sequentially pair the risk level labels at different time points in the dynamic risk level classification interface with the standardized state data sequence at the corresponding time points, to construct a training sample set containing input-output mapping relationships, and by substituting the training sample set into the prediction driving model constructed based on the time series supervised learning method, the prediction driving model inference operation is performed to obtain a future state-risk prediction sequence covering multiple future time points.

[0110] It needs to be explained that the prediction driving model constructed based on the time series supervised learning method is to organize the risk level label sequence and the standardized state data sequence in the dynamic risk level classification interface in the time dimension as a training sample pair set with event order characteristics, then a supervised learning structure with time series modeling capability (such as long short-term memory network, gated recurrent unit network, or time convolution network, etc.) is used to take the standardized state data sequence in the training sample pair set as input to generate the risk level label at the corresponding time point, and by minimizing the loss function (such as cross-entropy loss or mean square error loss) between the predicted output and the true label, parameter update iteration is performed to gradually optimize the time series mapping capability of the prediction driving model.

[0111] Based on the future state-risk prediction sequence and the working condition parameter constraint set, a rolling window optimization method is adopted to set a fixed length time window to slide on the future state-risk prediction sequence, extract the future state-risk prediction sequence within the window, and combine the upper and lower limit ranges of various process parameters in the working condition parameter constraint set and the change rate constraint conditions to construct a parameter feasible region, then a target function expression is constructed in the parameter feasible region by a target driving method, and according to the risk performance of the state-risk mapping relationship under different parameter combinations, a feasible working condition parameter combination set within the current window period is derived, and finally by continuously rolling the window, a working condition parameter adjustment solution space covering the entire future prediction period is gradually accumulated.

[0112] It needs to be explained that the rolling window optimization method refers to dividing the future state-risk prediction sequence into multiple continuous or overlapping time periods in the time series prediction scenario, and independently performing parameter optimization in each window to realize continuous updating of the strategy over time.

[0113] The various types of process parameters in the working condition parameter constraint set include production equipment operating temperature, rolling speed, cooling rate, heating temperature, hydraulic pressure, feeding speed, lubricating oil supply amount, and key production variables related to material ratio and energy consumption control.

[0114] On the basis of the constructed working condition parameter adjustment solution space, each group of working condition parameter combinations in the solution space is evaluated and compared according to a preset optimization objective function, the objective function value corresponding to each group of working condition parameter combinations is calculated, and the optimal solution group that satisfies the limitation conditions of various types of process parameters in the working condition parameter constraint set is selected through the ordering of the objective function values, and finally the corresponding optimization production parameter adjustment instruction is obtained.

[0115] The formula for calculating the objective function value corresponding to each group of working condition parameter combinations is as follows,

[0116]

[0117] wherein, indicates the objective function value corresponding to each group of working condition parameter combinations, H indicates the total number of prediction time steps, p indicates the current prediction time period number, and indicates the risk score weight factor, indicates the prediction risk score value corresponding to time step p, Q indicates the number of working condition parameters, q indicates the number of working condition parameters, and indicates the risk score value of the qth working condition parameter at time step p, q indicates the adjustment cost coefficient of the qth working condition parameter, indicates the target reference value of the qth working condition parameter at time step p, q,p indicates the current value of the qth working condition parameter at time step p.

[0118] It should be explained that the preset optimization objective function is constructed by fusing historical production working condition data statistical characteristics, experience parameter reference values and risk level information in the future state-risk prediction sequence according to the risk level control and process stability requirements in the seamless steel pipe production process, and the optimization objective function is composed of a target-oriented loss term and a parameter deviation term, wherein the loss term reflects the risk level evaluation result of the future time period, the deviation term measures the difference between various types of working condition parameters and the target reference value, and the weight coefficients are respectively assigned to regulate the optimization bias.

[0119] In summary, the present application realizes the parameter adjustment strategy reasoning driven by the risk evolution trend by adopting the prediction driving method, generating the optimization production parameter adjustment instruction according to the dynamic risk level grading interface and the standardized state data sequence, constructs the adjustment instruction generation mechanism combining the future risk situation change and the current process state, thereby realizing the forward-looking intervention and risk control of the seamless steel pipe production process, and finally achieving the beneficial effects of guaranteeing the production continuity and reducing the process failure rate.

[0120] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not to limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A seamless tube production line real-time monitoring system, characterized by: The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system.

2. The seamless tube production line real-time monitoring system of claim 1, wherein: The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system.

3. The seamless tube production line real-time monitoring system of claim 2, wherein: The application relates to a production line monitoring method and system.

4. The seamless tube mill real-time monitoring system of claim 1, wherein: The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring method and system. The application relates to a production line monitoring 5. The seamless tube mill real-time monitoring system of claim 1, wherein: The step of comparing the dynamic soft threshold boundary with the real-time observation value of the corresponding time point in the standardized state data sequence, extracting preliminary abnormal markers, and screening false abnormal markers by using a pseudo abnormality screening mechanism to generate a credible abnormality layer is as follows, The real-time observation value of each channel in the standardized state data sequence is compared with the upper and lower limits of the dynamic soft threshold boundary, and a preliminary abnormal marker set is generated by summarizing and generating; The preliminary abnormal marker set is screened by a pseudo abnormality determination to remove false abnormal markers caused by non-real factors, and a credible abnormality marker set is generated after screening and correction; According to the credible abnormality marker set, the time point and channel number of the marked abnormality in the standardized state data sequence are located, the credible abnormality layer construction field set is extracted from each group of time points and channel numbers, and the credible abnormality layer is constructed in a structured data form.

6. The seamless tube production line real-time monitoring system of claim 5, wherein: The pseudo abnormality determination process refers to identifying and removing false abnormal marker caused by non-real state changes in the preliminary abnormal marker set.

7. The seamless tube mill real-time monitoring system of claim 5, wherein: The field structure of the credible abnormality layer construction field set includes channel number, timestamp, standardized state variable value, energy mapping index value, dynamic soft threshold boundary upper and lower limit, and state variable corresponding energy feature dimension contribution information.

8. The seamless tube mill real-time monitoring system of claim 1, wherein: The step of dynamically generating a production indicator rendering layer interface based on the standardized state data sequence and the preliminary abnormal marker, and dynamically generating a risk perception interface based on the fusion calculation of the standardized state data sequence and the preliminary abnormal marker is as follows, Based on the state values of all channels and time points in the standardized state data sequence and the preliminary abnormal marker, an indicator rendering matrix is formed; Based on the indicator rendering matrix, a production indicator rendering layer interface is dynamically generated by using layer rendering method; The risk score sequence is calculated by a weighted fusion method by combining the abnormal distribution characteristics of the preliminary abnormal marker, and the risk perception interface is generated by fusing the risk score sequence with the production indicator rendering layer interface.

9. The seamless tube mill real-time monitoring system of claim 1, wherein: The step of fusing the credible abnormality layer into the risk perception interface to generate a dynamic risk level classification interface combined with the production indicator rendering layer interface is as follows, The abnormal positioning information set is extracted from the credible abnormality layer and integrated to build an abnormal channel time distribution table; Based on the abnormal channel time distribution table, the risk score data in the risk perception interface, and the abnormal positioning information in the credible abnormality layer, an abnormal propagation path graph is constructed, and then the risk level threshold interval is dynamically divided according to the correlation strength of the abnormal propagation path graph and the risk score data; The risk level threshold interval is fused with the production indicator rendering layer interface by using risk level graphical coding method to obtain a rendering layer interface with superimposed risk level labels; Based on the rendering layer interface with superimposed risk level labels, the rendering layer interface is dynamically updated by time series driving method to form a dynamic risk level classification interface.

10. The seamless tube mill real-time monitoring system of claim 1, wherein: The step of generating an optimized production parameter adjustment instruction based on the dynamic risk level classification interface and the standardized state data sequence is as follows, A prediction driving model is constructed based on the dynamic risk level grading interface and the standardized state data sequence, and a future state-risk prediction sequence is generated by reasoning; An optimal adjustment strategy is solved for the future state-risk prediction sequence and a working condition parameter constraint set, and a working condition parameter adjustment solution space is constructed; An optimization production parameter adjustment instruction is obtained by optimizing the working condition parameter adjustment solution space according to a preset optimization objective function.

Images