Method and system for constructing high-quality dataset for non-ferrous metal production process

Abstract

The application provides a non-ferrous metal production process-oriented high-quality data set construction method and system, and belongs to the technical field of industrial big data processing.The application extracts natural physical anchor points from the cooperative sudden change of furnace temperature and pressure to replace the absolute clock, fundamentally eliminates the time sequence error caused by clock drift and communication delay, realizes robust alignment of multi-source heterogeneous data, uses synchronous jump states of vibration and current signals to construct survival markers, can automatically distinguish real material events from sensor suspected failure, and avoids the interference of data noise on event detection in harsh working conditions; and the composite tracking code composed of sampling positions, container numbers and operation sequences is used to realize full-chain physical tracing from sampling to testing and forced back-hanging of test results, so that the high-precision correlation of process parameters and quality indexes is ensured.

Description

Technical Field

[0001] This invention belongs to the field of industrial big data processing technology, specifically relating to a method and system for constructing high-quality datasets for non-ferrous metal production processes. Background Technology

[0002] Non-ferrous metal smelting, especially aluminum electrolysis, is a typical complex industrial system characterized by high-dimensional nonlinearity and strong multiphase coupling, involving the interaction of multiple physical fields such as mass transfer, heat transfer, and electrochemical reactions. The internal operating parameters of the aluminum electrolysis cell exhibit high spatiotemporal coupling and are subject to intermittent process disturbances such as raw material fluctuations, electrode replacement, and aluminum feeding / discharging, resulting in significant dynamic characteristics in the process response. With the rapid development of the Industrial Internet of Things (IIoT), multi-source sensor networks, and artificial intelligence (AI) technologies, modern aluminum electrolysis production systems have deployed heterogeneous sensing devices on the electrolysis cell, including thermocouple arrays, pressure transmitters, anode current sensors, vibration accelerometers, and industrial cameras. This enables automated real-time acquisition of multimodal parameters such as electrolyte temperature, alumina concentration, aluminum liquid level, cell voltage, anode current density, flue gas pressure, and shell-breaking vibration. Existing technologies have constructed a full-process digital system covering multi-source sensing, intelligent analysis, and precise control. In terms of multi-source data fusion models, intelligent identification of superheat and balanced control of regional alumina concentration have been achieved, significantly improving the intelligence level of electrolysis cell operation. However, the construction of high-quality datasets for aluminum electrolysis process optimization, abnormal operating condition early warning, and product quality prediction still faces core technical bottlenecks such as inconsistent time series of multi-source heterogeneous data, reliability degradation under strong sensor interference, and difficulty in accurately correlating sampling and testing results with process parameters.

[0003] For example, CN121233607A discloses a multi-source data fusion acquisition system for aluminum electrolytic cells. This system electrically connects a multi-source data acquisition module to multiple sensors and cameras on the electrolytic cell to achieve real-time acquisition of multi-dimensional data. It timestamps various sensor data using a unified time reference and fuses heterogeneous data based on these timestamps to eliminate data acquisition time differences and improve data consistency. This system can achieve high-frequency, multi-source fusion, and low-latency data acquisition to meet the higher requirements of real-time performance and intelligence in the Industrial Internet of Things (IIoT). However, this system relies entirely on a single absolute clock reference for data alignment. In production scenarios like aluminum electrolytic cells, where strong electromagnetic interference, temperature fluctuations, and mechanical vibrations coexist, the drift of crystal oscillators in acquisition cards from different sensor channels, communication network delays, and the uncertainty of application layer data arrival times make the pure clock-stamp alignment method prone to introducing non-negligible time errors. More importantly, this system only solves the problem of timestamp unification at the data acquisition level. It lacks effective data quality marking and authenticity verification methods for partial signal failures caused by sensor loosening, impact, or electromagnetic interference. Furthermore, the system lacks a precise backtracking and matching mechanism between the reconstructed data and the sampling test results, thus failing to provide a high-quality training dataset for anchoring the true values ​​of samples for subsequent quality prediction models.

[0004] CN118691165A discloses a method and system for quality monitoring and traceability of aluminum smelting products based on big data modeling. This method uses a sensor network to collect real-time data on key parameters such as temperature, pressure, component concentration, and cooling rate during the aluminum smelting process. The collected data is preprocessed to establish a big data model of the aluminum smelting products. Real-time detection and analysis of product quality are then performed, and traceability analysis is conducted based on historical data from the model to pinpoint quality defects to specific process steps, thereby optimizing key parameters. This method provides a systematic technical approach for product quality monitoring and data-driven quality analysis. However, the data traceability of this method relies on ex post-hoc correlation analysis of historical data in big data models, that is, tracing back key process factors affecting quality through statistical fitting or regression models. This traceability method is essentially an indirect inference rather than a physical tracking of the entire process of sampled materials from the production site to the laboratory. The correlation between sampling time and test results still relies on batch number or timestamp matching. In scenarios such as different containers at the same sampling point or multiple reuses of the same container, material confusion is likely to occur, making it difficult to ensure accurate anchoring between test results and production parameters. Furthermore, it is impossible to attach quality labels back to a unified process state sequence for supervised learning and training. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by this invention is that the existing technology has shortcomings in constructing high-quality datasets, which cannot simultaneously address aspects such as the alignment of multi-source heterogeneous industrial data, the judgment of data confidence under sensor failure scenarios, and the accurate binding of sampling and testing results with process parameters.

[0008] To address the aforementioned technical problems, the present invention provides the following technical solution: A method for constructing high-quality datasets for non-ferrous metal production processes includes: extracting natural physical anchor points from furnace temperature and pressure signals to obtain an anchor point sequence; replacing the absolute clock with the anchor point sequence, realigning the time axis by matching anchor points across data sources, and outputting aligned multi-source data segments; attaching a survival marker to each data segment, consisting of the synchronous change states of parallel, same-window vibration signals and current signals, where a single signal jump is marked as a suspected sensor failure and the original signal value of the sampling point is retained, while multiple synchronous signal jumps are marked as real material events; generating a composite tracking code for each batch of sampled materials, consisting of the sampling location, container number, and operation sequence, carrying the composite tracking code throughout the sampling to testing process, and forcibly attaching the test results back to the production parameters at the sampling time corresponding to the data segment.

[0009] In a preferred embodiment of the present invention, the process of generating the anchor point sequence includes: continuously acquiring the temperature value of the furnace thermocouple and the pressure value of the pressure transmitter; calculating the first-order difference of the temperature value and the normalized first-order difference of the pressure value; when the first-order difference of the temperature value is less than a preset negative temperature threshold and the normalized first-order difference of the pressure value is less than a preset negative pressure threshold, the corresponding sampling point is recorded as a sudden drop inflection point and marked as a feeding candidate; calculating the first-order difference of the temperature value within a fixed-length sampling window after each sudden drop inflection point, detecting the zero-crossing point where the first-order difference of the temperature value changes from negative to positive, marking the first zero-crossing point as a gradual rise inflection point and as a slag discharge candidate, and pairing the sudden drop inflection point with the gradual rise inflection point; if there is no zero-crossing point within the sampling window, the sudden drop inflection point is discarded; the normalized first-order difference of the pressure value is the first-order difference of the pressure value divided by the range of the pressure transmitter.

[0010] As a preferred embodiment of the present invention, the anchor point sequence is formed by: alternating all effective paired sudden drop abrupt change points and corresponding gradual rise inflection points in ascending order of sampling sequence number, and each anchor point records its own sampling sequence number, absolute timestamp and type label to form an anchor point sequence; wherein each feeding anchor point is followed by a slag discharge anchor point, and adjacent anchor point pairs are feeding followed by slag discharge or slag discharge followed by feeding.

[0011] As a preferred embodiment of the present invention, the output aligned multi-source data segment includes: sequentially extracting two adjacent anchor points from the anchor point sequence in chronological order, using the absolute timestamp of the previous anchor point as the interval start point and the absolute timestamp of the next anchor point as the interval end point to form independent intervals; linearly mapping the original sampling timestamps of all data sources within each independent interval to normalized pseudo-time coordinates, such that the interval start point corresponds to the pseudo-time zero point and the interval end point corresponds to the pseudo-time one point, and saving the mapping relationship between each original timestamp and the pseudo-time; for each data source, taking points on the pseudo-time axis at a fixed step size, and performing linear interpolation using the original sampling values ​​and the mapping relationship to obtain aligned data with equal pseudo-time intervals; splicing all independent intervals in chronological order to obtain the aligned multi-source data segment, while saving the start and end absolute timestamps and interval number of each independent interval.

[0012] As a preferred embodiment of the present invention, the construction of the survival marker includes: for each independent interval, extracting vibration sensor signals and current sensor signals, and calculating the unit time change rate of adjacent sampling points; when the vibration change rate is greater than a preset vibration jump threshold and the current change rate is less than or equal to the current jump threshold, or when the current change rate is greater than the current jump threshold and the vibration change rate is less than or equal to the vibration jump threshold, the corresponding sampling point is marked as a suspected sensor failure, and the original signal value of the sampling point is retained; when both the vibration change rate and the current change rate exceed their respective thresholds and the vibration change direction is the same as the current change direction, the corresponding sampling point is marked as a real material event; for each original sampling point, the number of the independent interval to which it belongs is determined according to the original timestamp of the original sampling point, the corresponding pseudo-time value is found through the mapping relationship of the independent interval, and the marker of the original sampling point is assigned to the row with the smallest absolute value of the difference between the pseudo-time and the pseudo-time value corresponding to the original sampling point in the same independent interval of the aligned multi-source data segment.

[0013] As a preferred embodiment of the present invention, the composite tracking code includes: during sampling, reading the sampling point number, scanning the QR code of the sampling container to obtain the container number, having the operator enter the current process sequence number, and combining them into a composite tracking code; writing the composite tracking code into the sampling record table, and simultaneously recording the absolute sampling time; pasting the composite tracking code in the form of a QR code onto the sampling container, and scanning the QR code during testing to automatically associate the test results with the composite tracking code; retrieving the absolute sampling time based on the composite tracking code, finding an independent interval containing the absolute sampling time, and if the absolute sampling time is exactly the common boundary of adjacent independent intervals, selecting the previous independent interval, calculating the pseudo-time value corresponding to the absolute sampling time using the start and end absolute timestamps of the independent intervals, and filtering out the row with the smallest absolute value of the difference between the pseudo-time and the pseudo-time value within the same independent interval from the data segment with survival markers, and writing the test results into the test result field of the row.

[0014] As a preferred embodiment of the present invention, the preceding independent interval is selected when the absolute sampling time is exactly the common boundary of adjacent independent intervals, and the mapping from the original sampling timestamp to the pseudo-time coordinate within the independent interval adopts a linear mapping relationship.

[0015] On the other hand, the present invention also provides a high-quality dataset construction system for non-ferrous metal production processes, including: an anchor point extraction module, which extracts natural physical anchor points from furnace temperature and pressure signals to obtain an anchor point sequence; The timing alignment module replaces the absolute clock with the anchor point sequence, matches anchor points across data sources to realign the time axis, and outputs the aligned multi-source data segments. The survival marker module adds a survival marker to each data segment, consisting of the synchronous change status of parallel vibration signals and current signals within the same window. A single signal jump marks the sensor as suspected failure and retains the original signal value of the sampling point, while multiple signal jumps simultaneously mark it as a real material event. The tracking and backlinking module generates a composite tracking code for each batch of sampled materials, consisting of the sampling location, container number, and operation sequence. This composite tracking code is carried throughout the entire process from sampling to testing, and the test results are forcibly backlinked to the production parameters corresponding to the sampling time in the data segment.

[0016] The beneficial effects of this invention are as follows, compared with the prior art: This invention fundamentally eliminates timing errors caused by clock drift and communication delays by extracting natural physical anchor points from the coordinated abrupt changes in furnace temperature and pressure to replace absolute clocks, achieving robust alignment of multi-source heterogeneous data. Simultaneously, it utilizes the synchronous transition states of vibration and current signals to construct survival markers, automatically distinguishing between real material events and suspected sensor failures, avoiding interference from data noise in event detection under harsh operating conditions. Furthermore, it employs a composite tracking code composed of sampling location, container number, and operation sequence to achieve full-chain physical traceability from sampling to testing and forced re-attachment of test results, ensuring high-precision correlation between process parameters and quality indicators. The dataset constructed by this invention features high temporal consistency, strong interpretability of event labels, and accurate quality anchoring, providing a high-quality data foundation for process optimization, fault diagnosis, and quality prediction models in non-ferrous metal smelting processes, significantly enhancing the value of industrial data assets. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method for constructing high-quality datasets for non-ferrous metal production processes as described in this invention.

[0018] Figure 2 This is a structural diagram of a high-quality dataset construction system for non-ferrous metal production processes as described in this invention.

[0019] Figure 3 A schematic diagram of the structure of an electronic device for implementing the method of constructing a high-quality dataset for non-ferrous metal production processes according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0021] like Figure 1 As shown, the method for constructing a high-quality dataset for non-ferrous metal production processes according to the present invention includes: S1: Extract natural physical anchor points from the furnace temperature and pressure signals to obtain the anchor point sequence.

[0022] This embodiment uses a large aluminum electrolysis cell as an example. During the aluminum electrolysis production process, periodically adding alumina to the cell (feeding) causes a sudden drop in cell temperature and negative pressure (or flue gas pressure); while tapping aluminum or discharging electrolyte (aluminum tapping) causes the cell temperature to slowly rise. To extract the natural physical anchors reflecting process events (feeding and aluminum tapping), the following steps are performed. Specifically: S1.1: Multiple thermocouples, such as cathode steel rod thermometers or sidewall thermometers, are installed inside the aluminum electrolysis cell. The output is converted into a 4-20mA current signal by a temperature transmitter and continuously acquired by the PLC system at a sampling frequency of 1Hz to obtain the temperature value sequence T(i), in °C. Simultaneously, a pressure transmitter with a range of -1kPa to +1kPa (i.e., a range of 2kPa) is installed in the flue gas duct above the electrolysis cell or inside the cell hood. This transmitter also acquires the pressure value sequence P(i), in kPa, at a sampling frequency of 1Hz. The sampling sequence i increments from 1, and the absolute timestamp corresponding to each sampling point is obtained by associating it with the PLC system clock, in UNIX timestamp format (unit: seconds).

[0023] The data collection process is continuous and does not depend on external triggers.

[0024] S1.2: For each sampling number i (i≥2), calculate the first-order difference of the temperature value ΔT(i)=T(i)-T(i-1), with the unit being ℃ / s. This difference reflects the rate of temperature change over time.

[0025] Simultaneously, the first-order difference of the pressure value ΔP(i) = P(i) - P(i-1), in kPa / s, is calculated, and then divided by the range of the pressure transmitter (in kPa) to obtain the normalized first-order difference ΔP. norm (i).

[0026] It should be noted that the purpose of normalization is to eliminate the influence of differences in the range of different electrolytic cells or different pressure transmitters, so as to make the threshold setting universal.

[0027] The preset negative threshold for temperature is -0.5℃ / s, and the preset negative threshold for pressure is -0.05. This condition is met when ΔT(i) < -0.5 and ΔP... norm When (i) < -0.05, it indicates that the cell temperature is rapidly decreasing and the pressure is also decreasing rapidly at the same time, which is a typical characteristic of adding cold alumina (feeding) to the electrolytic cell. At this time, the sampling point i is recorded as the sudden drop point and marked as a feeding candidate. The absolute timestamp of this sudden drop point is the physical anchor time of the feeding event.

[0028] It should be noted that the unit of the first-order temperature difference is ℃ / s, while the unit of the normalized first-order pressure difference is 1 / s, which is the proportion of pressure change per second relative to full scale. Although the units are different, they are determined independently using thresholds of -0.5℃ / s and -0.05 (dimensionless), without requiring physical consistency of the values. These thresholds are statistically calibrated based on the rate of temperature and pressure change caused by feeding events in historical data from aluminum electrolysis cells.

[0029] It should be noted that a material feeding candidate is only identified when both temperature and pressure drop simultaneously. A single signal change (e.g., temperature drops while pressure stabilizes) may be sensor noise or a local disturbance and is not considered a valid material feeding event. This dual-condition determination significantly reduces the false detection rate.

[0030] S1.3: For each marked sudden drop point (feeding candidate), take the next 50 consecutive sampling points as the detection window. If the window length exceeds the end of the data acquisition, take the data up to the end of the data.

[0031] Within this window, the first-order difference sequence ΔT(i) of the temperature values ​​is calculated. Then, starting from the first sampling point after the abrupt drop, a sequential scan is performed to find the first sampling point k that satisfies the following conditions: ΔT(k-1) < 0 and ΔT(k) ≥ 0. To eliminate false zero-crossings caused by high-frequency noise in the temperature signal, a moving average filter is applied to the temperature value sequence T(i) before detection. The window length is 3 sampling points, i.e., the average of the current point and the points before and after it, resulting in the filtered temperature sequence. Then based on Calculate the first-order difference In addition, a dead zone condition is set for zero-crossing detection: only when... After the value changes from negative to non-negative, at least two consecutive sampling points Only when all zero values ​​are greater than or equal to 0 can the zero-crossing point be considered a valid inflection point of gradual rise.

[0032] This condition indicates that the first-order difference of temperature changes from a negative value to a non-negative value. In practical applications, due to the precision of numerical values, the change from negative to positive or from negative to zero is usually detected. This point is the inflection point where the rate of temperature decrease changes from negative to zero and then back to positive, marking the point where the temperature stops decreasing and begins to rise. Correspondingly, after feeding, the electrolytic cell gradually restores thermal equilibrium and begins to produce aluminum or discharge electrolyte.

[0033] The first zero-crossing point detected is marked as a gradual rise inflection point and a candidate for aluminum tapping. The absolute timestamp of this gradual rise inflection point is the physical anchor point time of the aluminum tapping event. Simultaneously, a pairing relationship is established between this gradual rise inflection point and the corresponding abrupt drop inflection point in S1.2, meaning one feeding candidate uniquely corresponds to one subsequent aluminum tapping candidate. If no zero-crossing point where the first-order difference changes from negative to positive is detected within the window, it is considered that no valid aluminum tapping event has occurred after the abrupt drop inflection point. In this case, the abrupt drop inflection point is recorded separately as an isolated feeding candidate, not included in the main sequence of the anchor point sequence, but its information is stored in an auxiliary table for subsequent manual verification or model compensation.

[0034] It is important to emphasize that this step uses the first-order difference zero-crossing point instead of the second-order difference zero-crossing point. The second-order difference zero-crossing point corresponds to the extreme point of the first-order difference (the inflection point of the decreasing rate), at which point the temperature is still decreasing and is not the true starting point of the rebound. Using the first-order difference zero-crossing point from negative to positive can accurately capture the instant when the temperature changes from decreasing to increasing, which closely matches the physical process of aluminum starting to be produced from the electrolytic cell.

[0035] S1.4: Arrange all valid paired sudden drop points and corresponding gradual rise inflection points alternately in ascending order of sampling number.

[0036] Since the sampling sequence number of the abrupt drop point in each pair is necessarily less than its corresponding gradual rise inflection point, and different pairs do not overlap, the resulting sequence naturally forms an alternating order of feeding-aluminum tapping-feeding-aluminum tapping… It should be noted that, because feeding and aluminum tapping may not be strictly one-to-one alternations during the process—for example, two consecutive feedings followed by one aluminum tapping—this step requires each feeding candidate to be paired with a subsequent aluminum tapping candidate. Unpaired feedings are discarded. Therefore, adjacent anchor point pairs in the final sequence may have two cases: one is feeding followed by aluminum tapping, i.e., from the same pair; the other is aluminum tapping followed by feeding, i.e., from the aluminum tapping of the previous pair and the feeding of the next pair. Both cases are allowed, but each feeding anchor point must be followed by an aluminum tapping anchor point.

[0037] Each anchor point records the following three fields: sampling sequence number i, absolute timestamp, and type label. Arranging these anchor points in chronological order yields anchor point sequence A.

[0038] It can be seen that the anchor point sequence A is generated entirely based on the physical signal changes of the aluminum electrolysis cell itself, without relying on external clock synchronization signals or artificial markers, and has a natural anti-drift capability. Compared with a time axis using an absolute clock, this anchor point sequence divides the continuous production process into several process intervals defined by feeding and aluminum tapping events, and the signal changes within each interval have similar physical mechanisms.

[0039] S2: Replace the absolute clock with the anchor point sequence, match the anchor points across data sources to re-align the time axis, and output the aligned multi-source data segment.

[0040] It should be noted that in step S1, an anchor point sequence A, consisting of alternating feeding and aluminum exit anchor points, was extracted from the temperature and pressure signals of the aluminum electrolysis cell. Each anchor point carries an absolute timestamp. However, vibration sensor signals and current sensor signals are also simultaneously acquired during the aluminum electrolysis production process. These data sources, along with the temperature and pressure signals, come from different acquisition cards or have different sampling frequencies, resulting in time offsets and nonlinear drifts between them. Absolute clocks cannot accurately align the event times of different data sources due to equipment clock drift, communication delays, and other reasons. This embodiment uses anchor point sequence A to replace the absolute clock, realigning the time axes of different data sources, making the data within the same process stage comparable on a pseudo-time coordinate. The specific operation is as follows:

[0041] S2.1: Take two adjacent anchor points from the anchor point sequence A in chronological order, and denote the previous anchor point as... The next anchor point is Previously an anchor point absolute timestamp The starting point of the interval, followed by the next anchor point. absolute timestamp This serves as the endpoint of the interval, forming an independent interval. Since the anchor points in anchor point sequence A are arranged in ascending order of sampling number,... It must be less than .

[0042] It should be noted that adjacent anchor point pairs in anchor point sequence A may be of two types: one is the feeding anchor point and the aluminum tapping anchor point generated by the same effective pairing, i.e., feeding → aluminum tapping; the other is the aluminum tapping → feeding interval formed by the previous pairing of aluminum tapping anchor points and the next pairing of feeding anchor points. Both types of intervals are treated independently, each corresponding to a complete process sub-stage of the aluminum electrolysis cell: the feeding → aluminum tapping interval corresponds to the dissolution reaction process from the addition of alumina to the start of aluminum tapping, while the aluminum tapping → feeding interval corresponds to the waiting or thermal equilibrium process from the end of aluminum tapping to the start of the next feeding. Treating the two types of intervals separately avoids mixing different physical processes in the same reference frame.

[0043] S2.2: For an independent interval, its starting absolute timestamp is... The absolute timestamp of the endpoint is Within this interval, the original sampling timestamps t of all data sources lie within the closed interval. Inside.

[0044] To eliminate the absolute duration differences between intervals, t is linearly mapped to a pseudo-time coordinate. Make the starting point of the interval correspond =0, the endpoint of the interval corresponds to =1. The mapping formula is: ; Because this embodiment uses a linear mapping The values ​​are uniformly distributed within the interval [0,1], independent of the scaling ratio of the absolute duration. For example, a 10-second interval and a 100-second interval are both compressed or stretched to the same length on the pseudo-time coordinate, allowing direct comparison of data from the same process stage within different intervals. Simultaneously, for each original sampling timestamp t within this interval, its mapped value is saved. The values ​​form a mapping table. ( (This refers to the numbering of independent intervals). The mapping table uses the original timestamp t as the key and... Since the sampling frequencies of vibration sensors, current sensors, temperature sensors, and pressure sensors are different, this embodiment establishes an independent mapping table for each sensor channel. For any original sampling point (including all sensors), its timestamp can be found in the mapping table of its respective sensor channel. If the value exactly matches a key in the mapping table, then the value is directly retrieved. Otherwise, take the value corresponding to the timestamp closest to the timestamp in the mapping table. value.

[0045] It should be noted that using a linear mapping instead of an absolute clock eliminates the time axis scaling differences caused by varying absolute durations across different intervals. This ensures that the seconds following material feeding and the seconds before aluminum tapping have consistent positional significance in pseudo-time, greatly facilitating subsequent data fusion and pattern recognition, and also saving the mapping table. It provides bidirectional query capabilities, allowing you to retrieve pseudo-time from the original time, or conversely (through interpolation) to approximate the original time from the pseudo-time.

[0046] S2.3: For each independent interval, a fixed pseudo-time step is determined. As an optional implementation, each interval is uniformly divided into 100 equally spaced pseudo-time sub-intervals, generating a total of 101 pseudo-time points with a step size of 0.01. The number of divisions can be configured according to the actual signal change rate; for example, 200 divisions can be used for rapidly changing processes, while 50 divisions can be used for slowly changing processes. The default value for the number of divisions is 100, and it can be adjusted via a configuration file, thus obtaining a fixed... A set of values, such as {0, 0.01, 0.02, 0.03, …, 1.00}.

[0047] For each data source, such as temperature, pressure, vibration, and current, it is necessary to obtain each The interpolated value corresponding to the target value. Specifically, for a given target... The value is first obtained using the saved mapping table. Find the original timestamp t within this interval. The value sequence, and the original measurements corresponding to these original timestamps t; then, in Find the target value in the value sequence Two adjacent values Value, that is, a value less than or equal to the target. A target greater than or equal to According to these two Linear interpolation is performed on the value and its corresponding original measurement value to calculate the target. The interpolation result at the value. For example, if the target =0.25, there exists within the interval 1 = 0.24 corresponds to the original temperature value T1. Since 2 = 0.27 corresponds to the original temperature value T2, the interpolation result = T1 + (0.25 - 0.24) / (0.27 - 0.24) × (T2 - T1). For =0 and =1. For the two boundary points, directly take the original values ​​at the start and end anchor points of the interval.

[0048] Furthermore, after performing the above interpolation on all data sources separately, each fixed At each value, a set of interpolated data was obtained: interpolated temperature values. interpolated pressure value Interpolated vibration values Interpolated current value .

[0049] It is important to emphasize that linear interpolation is suitable for scenarios where signals change approximately linearly over a short period of time. In this embodiment, since the process signals (temperature, pressure, etc.) in the aluminum electrolysis cell change relatively slowly and the sampling frequency is high (1Hz), the signal fluctuations between adjacent sampling points are small, and the accuracy of linear interpolation is sufficient to meet the requirements of dataset construction. If the signal changes drastically, spline interpolation can also be used, but this will increase computational complexity. This embodiment chooses linear interpolation to balance accuracy and efficiency.

[0050] S2.4: Concatenate the processed independent intervals sequentially according to the time order in anchor sequence A. Within each independent interval, the data rows are arranged according to... The values ​​are arranged in order from 0 to 1. During concatenation, data rows from different intervals are arranged directly and continuously without overlap or smoothing, because different intervals correspond to different physical stages and should not be mixed.

[0051] After concatenation, the aligned multi-source data segment D is obtained. D is a tabular data structure, and each row contains the following fields: pseudo-time coordinates. Interval number j, interpolated temperature value interpolated pressure value Interpolated vibration values Interpolated current value The interval number j is used to distinguish different independent intervals and avoid different intervals having the same... The rows of values ​​are mixed up; at the same time, a separate interval boundary mapping table is maintained, and each row of this table records the following information for an independent interval: interval number j, absolute timestamp of the interval start point. Absolute timestamp of the endpoint of the interval This mapping table will be used in subsequent steps to backmap the absolute sampling time to pseudo-time coordinates.

[0052] Each row in dataset D represents a process state point aligned in pseudo-time, and the values ​​from different data sources are in the same... The values ​​are comparable.

[0053] S3: Add a survival marker to each data segment, consisting of the synchronous change status of parallel vibration signals and current signals within the same window. A single signal change is marked as a suspected sensor failure and the original signal value of the sampling point is retained. Multiple synchronous signal changes are marked as a real material event.

[0054] It should be noted that the environment at the aluminum electrolysis cell site is harsh. Anode vibration sensors, accelerometers in the shell-breaking device, and current transformers are susceptible to strong magnetic fields, high temperatures, dust, and mechanical impacts, leading to loosening, interference, or localized malfunctions. This results in the collected signals failing to accurately reflect the material's movement. Indiscriminately treating all signal jumps as genuine process events will introduce significant noise and degrade the dataset quality. It is necessary to utilize the physical coupling characteristics of vibration and current signals—genuine material events, such as alumina impact during feeding, anode lifting and lowering during aluminum tapping, and shell-breaking actions, cause synchronous jumps in both vibration and current, while a single sensor failure typically only causes a jump in its own signal. By detecting the synchronous changes in both, a survival flag is added to each data segment, thus distinguishing between valid events and suspected sensor failures. The specific operation is as follows: S3.1: In the raw data before alignment, the vibration sensor and the current sensor may have different sampling frequencies. In this embodiment, the vibration sensor (accelerometer) installed on the anode mechanism or shell-breaking device of the aluminum electrolysis cell has a sampling frequency of 100Hz, and the current transformer of the motor driving the anode lifting or aluminum tapping has a sampling frequency of 50Hz. Since the timestamps of the two sensors have been unified to the same clock reference, the rate of change of their respective signals on the time axis can be calculated separately.

[0055] For vibration sensor signals The time interval between adjacent sampling points is denoted as (Typically 0.01 seconds, corresponding to 100Hz). For the i-th sampling point (i≥2), calculate the absolute value of the vibration rate of change: ; The unit is g / s (g is the acceleration due to gravity, 1g = 9.8 m / s²). This value represents the degree of drastic change in the amplitude of vibration per unit time.

[0056] For current sensor signals The time interval between adjacent sampling points is denoted as Typically, this is 0.02 seconds, corresponding to 50Hz. For the i-th sampling point (i≥2), calculate the absolute value of the rate of change of current: ; The unit is A / s. This value represents the degree of drastic change in current per unit time.

[0057] At the same time, calculate separately plus and minus signs and The plus or minus sign is used to determine whether the changes have the same sign in subsequent tests.

[0058] It should be noted that the original sampled signal is used here instead of the aligned and interpolated signal because transition detection requires the time resolution of the original signal, and interpolation may smooth out instantaneous transition features. This step is performed within the original time window, and then the markers are mapped to the aligned data segment D using the mapping relationship before alignment.

[0059] S3.2: Based on historical data statistics under normal operating conditions of the aluminum electrolysis cell, the vibration jump threshold is set to 5 g / s, and the current jump threshold is set to 30 A / s. These thresholds can be calibrated on-site according to different cell types or different operating conditions. The values ​​given in this embodiment are only examples.

[0060] For each original sampling point, the following determination is made: The determination condition for a single signal transition is as follows: Condition A: > 5 and ≤30, meaning the vibration rate of change exceeds the threshold while the current rate of change does not; Condition B: >30 and ≤5 means that the rate of change of current exceeds the threshold while the rate of change of vibration does not exceed the threshold.

[0061] Sampling points that meet either condition A or condition B are marked as potentially faulty sensors. Simultaneously, the original vibration values ​​of these sampling points are retained. and the original current value No smoothing or replacement is performed. The purpose of this marker is to remind data users that a single signal jump at that moment is likely due to sensor malfunction, poor wiring contact, electromagnetic interference, or strong magnetic field effects, rather than actual material movement.

[0062] Synchronous transition of multiple signals requires the simultaneous fulfillment of the following three sub-conditions: > 5; >30; and Same sign means that both increase or decrease at the same time.

[0063] Sampling points that meet all the above conditions are marked as real material events. This marking is used to confirm that physical processes such as feeding impact, anode lifting, shell breaking, or aluminum tapping actually occurred at that moment, and serves as a cross-validation and supplement to the anchor point events extracted based on temperature and pressure in S1. If neither the single-signal transition condition nor the multi-signal synchronous transition condition is met, it is marked as no event.

[0064] It is important to emphasize that the determination of real material events requires that vibration and current change synchronously and in the same direction. This utilizes physical causality, namely, the mechanical impact when alumina is added to the electrolytic cell will simultaneously cause the cell shell to vibrate (increased vibration) and the anode current to fluctuate (current change). When aluminum is tapped, the anode rising and falling will also simultaneously change the motor load current and mechanical vibration, and the directions of change of both are consistent (e.g., increasing or decreasing simultaneously). However, a single sensor failure (such as a loose vibration sensor) cannot cause a synchronous change in the other signal. This mechanism naturally quantifies the credibility of the event.

[0065] As can be seen, this invention constructs a survival marker by synchronously changing vibration and current states, which can automatically distinguish between real material events and suspected sensor failures in aluminum electrolysis cells without human intervention, greatly reducing the manual cost of subsequent data cleaning. At the same time, it preserves the original traces of suspected sensor failure markers, avoiding the erroneous deletion of abnormal data that may contain valid information (such as extremely rare real single-signal events), providing data users with a flexible filtering strategy.

[0066] S3.3: Since the above markings are made on the original sampling points within the original time window, while each row in the aligned multi-source data segment D corresponds to an interpolation point on a fixed pseudo-time step, the two need to be correlated.

[0067] Specifically, for each marked original sampling point, the independent interval number j to which the sampling point belongs is first determined. The determination includes: the original timestamp t of the sampling point being located within that interval. and Between (including boundaries). Since each independent interval is sequential and non-overlapping in time, each original sampling point must belong to a unique independent interval; using the mapping table of this independent interval... Find the corresponding pseudo-time based on the original timestamp t. Value. Due to the mapping table It stores the timestamp corresponding to each original sampling. The value can be directly obtained for that sampling point. Value. Since the sampling timestamps of different sensors usually do not coincide precisely, the nearest neighbor matching strategy is mainly used in practice, that is, in the mapping table. Find the original timestamp that is closest to the original sampling timestamp t, and take the corresponding one. The smaller of two equally spaced neighboring timestamps is used. The value, an exact match, only occurs in very rare cases, such as when two sensors share the same sampling clock; After setting the value, in the aligned multi-source data segment D, filter out all rows with interval numbers equal to j, and search within these rows. The value obtained above The row with the smallest absolute difference between values ​​is selected. The event label field of this row is assigned a label type, such as "Sensor Suspected Failure," "Real Material Event," or "No Event." A priority merging rule is used, where priority from highest to lowest is: Real Material Event > Sensor Suspected Failure > No Event. If multiple original sampling points are mapped to the same row, the event label for that row is the one with the highest priority. That is, if at least one original sampling point is labeled as a Real Material Event, the row is considered a Real Material Event; otherwise, if at least one sensor is suspected of failure, the row is considered a Sensor Suspected Failure; otherwise, it is considered No Event. This merging rule ensures that event labels in the data segment do not lose critical information due to downsampling.

[0068] After mapping all the original sampling points, a data segment with liveness markers is generated. .in, In addition to the existing fields in D, an extra column of event labels will be added.

[0069] S4: Generate a composite tracking code for each batch of sampled materials, consisting of the sampling location, container number, and operation sequence. Carry the composite tracking code throughout the entire process from sampling to testing, and forcibly attach the test results back to the production parameters at the sampling time corresponding to the data segment.

[0070] It should be noted that during the electrolysis production process, the produced molten aluminum or electrolyte needs to be sampled and tested, and the test results are correlated with production parameters to establish a process-quality relationship model. Traditionally, only the absolute clock is recorded at the sampling time, and test results are entered manually or associated with batch numbers. However, due to differences in sampling locations, mixed containers, and chaotic operation sequences, test results often fail to accurately correspond to specific production parameter segments. This step generates a composite tracking code to achieve full-chain tracking from sampling to testing. Using the interval boundary mapping table and pseudo-time mapping relationship saved in the previous steps, the test results are forcibly appended to the row in the data segment corresponding to the sampling time, ensuring that each test result is accurately bound to its corresponding process state. Specifically: S4.1: When on-site operators take samples from a fixed sampling point in the aluminum electrolysis cell, such as the aluminum liquid discharge port or the electrolyte sampling port, they first read the sampling point number from a pre-installed barcode reader or manual input device at the sampling point. In this embodiment, the sampling point number uses a 3-digit alphanumeric combination, for example, ALP represents the aluminum liquid discharge port (Aluminum Pouring), and ELP represents the electrolyte discharge port (Electrolyte Pouring). This number is physically fixed near the sampling point to avoid human error in selection.

[0071] Furthermore, the operator uses a handheld industrial tablet to scan the sampling containers that have been pre-affixed with QR codes. The QR code on the container has been written with a unique container number upon entry into the warehouse, such as C20231001-001. After scanning, the tablet automatically retrieves the container number.

[0072] Furthermore, the operator selects the current process number from the process drop-down menu on the tablet. Aluminum electrolysis production is usually carried out by shift or aluminum output sequence, and the process number uses a two-digit code. For example, 01 indicates the first aluminum output sampling, and 02 indicates the second aluminum output sampling or sampling after electrode change. If there is no specific process division, the default is 00.

[0073] The three fields mentioned above are combined in a fixed order to form a composite trace code, for example, ALP-C20231001-001-01. The structure of this composite trace code ensures uniqueness, meaning that different combinations of sampling points, containers, and processes can uniquely identify a single sampling event.

[0074] As can be seen, the composite tracking code integrates the sampling location, container identification, and operation sequence, overcoming the ambiguity caused by the traditional use of only batch number or only timestamp. Even if multiple sampling points are sampled at the same time or the same container is reused (after cleaning), the composite tracking code can accurately distinguish each batch of materials, providing an unalterable identifier for subsequent test results.

[0075] S4.2: Simultaneously with the generation of the composite tracking code, the tablet computer reads the current clock reading of the system and records it as the sampling absolute time. .Should Use the same clock reference as the absolute timestamp recorded at the anchor point in step S1.4 above to ensure time alignment. Combine the composite trace code and sample the absolute time. Optional field information (such as operator ID and electrolytic cell number) is also written to the sampling record table in the database. The writing operation is completed in real-time at the sampling site via wireless network, avoiding the loss of paper records or input errors.

[0076] S4.3: After filling the sampling container (sampling spoon or sample mold) with molten aluminum or electrolyte, the on-site operator uses a portable label printer to print a QR code label with the same Trace content and firmly affixes it to the outer wall of the sampling container or sample box. The QR code encoding content is the Trace string, for example, ALP-C20231001-001-01. The container is then sent to the laboratory.

[0077] Upon receiving the container, laboratory staff first scan the QR code on the outer wall of the container using a QR code scanner; the system automatically reads the Trace string. Next, the laboratory personnel perform sample pretreatment according to standard procedures and use analytical instruments to determine the chemical composition or electrolyte molecular ratio. The workstation software of the analytical instrument includes a sample ID input box. Laboratory personnel manually enter the scanned Trace string or automatically fill it into this input box via serial port, establishing a one-to-one binding between the test result and the Trace.

[0078] After the test is completed, the system writes the test results to the test result table in the database. The primary key of this table is also Trace, and the test completion timestamp is recorded. The composite tracking code enables the entire process from sampling to testing to be carried.

[0079] S4.4: When it is necessary to associate test results with a specific row in the production parameter data segment, execute the following reverse mapping process.

[0080] First, the absolute sampling time is obtained by querying the sampling record table in the database based on the composite tracking code Trace. (Unit: seconds, floating-point number); Traverse the interval boundary mapping table, where the search condition is: ≤ ≤ .like Strictly greater than a certain interval And strictly less than Then the interval is the only one containing The interval. If exactly equal to a certain interval In this embodiment, the preceding interval (i.e., the one with the smaller interval number) is explicitly selected. This boundary processing rule ensures consistency; even if the sampling time happens to fall on the anchor point (such as the end of aluminum extraction), it is still classified as an interval before the anchor point, because the anchor point itself represents the instant the event is completed, while sampling usually occurs after the event. It should be noted that the reason for selecting the preceding interval is because... The corresponding anchor point time (e.g., the moment aluminum tapping is completed) is the instant the event ends. Sampling usually occurs after the event ends, therefore assigning it to the interval following the event, i.e., the preceding interval, is more consistent with actual process conditions. The calculated value at this time... =1, and matched within the previous interval. The row with =1 actually represents the end state of the interval.

[0081] After determining the sampling time After identifying the independent intervals, calculate using the linear mapping relationship. Corresponding pseudo-time value Specifically, the calculation process is as follows: This equals the difference between the absolute sampling time and the absolute timestamp of the interval's start point, divided by the difference between the absolute timestamp of the interval's end point and the absolute timestamp of the interval's start point. Because the absolute sampling time... It may not completely coincide with any original sampling timestamp; this embodiment directly uses the above linear mapping formula for calculation. without relying on mapping tables This is because of the mapping table. It stores the mapping relationship of the original sampling points, and the linear mapping formula itself is a continuous function, which can be directly calculated for any t. .in, The value range of is [0, 1]. If Exactly equal to ,but =0; if equal ,but =1. The boundary processing rules in this embodiment will be... = The situation falls under the previous interval, therefore, in actual reversal, The case where =1 only occurs at the end of the previous interval.

[0082] Furthermore, in data segments with liveness markers In the text, each row already contains the interval number j and the pseudo-time. (Discrete values ​​with a step size of 0.01). First, filter using interval number j to obtain all rows belonging to that independent interval. Then, calculate the value for each row within these rows. Value and Find the row with the smallest absolute difference between the rows. For example, if... =0.256, then Within this interval There are two adjacent rows with values ​​of 0.25 and 0.26, with differences of 0.006 and 0.004 respectively, therefore, [the value is selected]. The row with a value of 0.26 (the difference is even smaller). If Exactly equal to a certain If the value is 0.25, then that row is selected. After finding the target row, the test result field of that row is assigned the test result retrieved from the database. If the row already has a test result, for example, multiple samples corresponding to the same pseudo-time row, or due to duplicate data entry causing a duplicate upload, then the latest overwrite strategy is used: the new test result overwrites the old value, and a test update timestamp field is appended to the row to record the time of this overwrite. This strategy ensures that each pseudo-time row in the final dataset retains at most one test result, and it is always the latest.

[0083] Finally, a complete production parameter data table with test results is generated. This table includes aligned multi-source production parameters, event tags, and test results at the corresponding sampling time. Each row represents a pseudo-time-aligned process state point, and if a corresponding sampling test exists for that state point, a quality indicator is attached.

[0084] Optional, if based on No matching interval boundary map could be found. ≤ ≤ The interval, for example, due to clock asynchrony. Exceeding the earliest or latest anchor point, or If the error occurs within an invalid time period before the start of the production process, the system records an anomaly log and temporarily stores the test results corresponding to the composite tracking code in an unrecovered table, awaiting manual verification or adjustment of the clock reference before re-encountering. This anomaly handling mechanism ensures the integrity of the dataset and prevents the loss of test results due to extreme circumstances.

[0085] As can be seen, the embodiments of the present invention realize the full-process tracking of aluminum electrolysis sampling materials by generating composite tracking codes, realize the accurate conversion from absolute sampling time to pseudo-time coordinates through interval boundary mapping table and linear mapping formula, and write the test results into the corresponding data segment rows through nearest neighbor matching. The final output dataset has a three-in-one structure of process parameters, event markers and quality indicators, providing a high-quality data foundation for process optimization, quality prediction and fault diagnosis in aluminum electrolysis production.

[0086] like Figure 2 As shown, the present invention also provides a high-quality dataset construction system for non-ferrous metal production processes, comprising: The anchor point extraction module extracts natural physical anchor points from furnace temperature and pressure signals to obtain an anchor point sequence. The timing alignment module replaces the absolute clock with the anchor point sequence, matches anchor points across data sources to realign the time axis, and outputs the aligned multi-source data segments. The survival marker module adds a survival marker to each data segment, consisting of the synchronous change status of parallel vibration signals and current signals within the same window. A single signal jump marks the sensor as suspected failure and retains the original signal value of the sampling point, while multiple signal jumps simultaneously mark it as a real material event. The tracking and backlinking module generates a composite tracking code for each batch of sampled materials, consisting of the sampling location, container number, and operation sequence. This composite tracking code is carried throughout the entire process from sampling to testing, and the test results are forcibly backlinked to the production parameters corresponding to the sampling time in the data segment.

[0087] The high-quality dataset construction device for non-ferrous metal production processes provided in this embodiment of the invention can execute the high-quality dataset construction method for non-ferrous metal production processes provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0088] Figure 3 This is a schematic diagram of an electronic device for implementing a method for constructing high-quality datasets for non-ferrous metal production processes, as described in an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptops, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0089] like Figure 3As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0090] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0091] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as methods for constructing high-quality datasets for non-ferrous metal production processes.

[0092] In some embodiments, the method for constructing a high-quality dataset for a non-ferrous metal production process can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for constructing a high-quality dataset for a non-ferrous metal production process described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the method for constructing a high-quality dataset for a non-ferrous metal production process by any other suitable means (e.g., by means of firmware).

[0093] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0094] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0095] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0096] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0097] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0098] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0099] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0100] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for constructing high-quality datasets for non-ferrous metal production processes, characterized in that, include: Natural physical anchor points are extracted from furnace temperature and pressure signals to obtain an anchor point sequence; Based on the anchor point sequence, the absolute clock is replaced, and the anchor points are matched across data sources to re-align the time axis, and the aligned multi-source data segments are output. Each data segment is appended with a survival marker consisting of the synchronous change states of parallel vibration signals and current signals within the same window. A single signal jump is marked as a suspected sensor failure and the original signal value of the sampling point is retained. Multiple synchronous signal jumps are marked as real material events. For each batch of sampled materials, a composite tracking code is generated, which consists of the sampling location, container number, and operation sequence. The composite tracking code is carried throughout the entire process from sampling to testing, and the test results are forcibly linked back to the production parameters at the sampling time corresponding to the data segment. The process of generating the anchor point sequence includes: continuously acquiring the temperature values ​​of the furnace thermocouples and the pressure values ​​of the pressure transmitters; calculating the first-order difference of the temperature values ​​and the normalized first-order difference of the pressure values; when the first-order difference of the temperature values ​​is less than a preset negative temperature threshold and the normalized first-order difference of the pressure values ​​is less than a preset negative pressure threshold, the corresponding sampling point is recorded as a sudden drop inflection point and marked as a feeding candidate; calculating the first-order difference of the temperature values ​​within a fixed-length sampling window after each sudden drop inflection point, detecting the zero-crossing point where the first-order difference of the temperature values ​​changes from negative to positive, marking the first zero-crossing point as a gradual rise inflection point and as a slag discharge candidate, and pairing the sudden drop inflection point with the gradual rise inflection point; if there is no zero-crossing point within the sampling window, the sudden drop inflection point is discarded; the normalized first-order difference of the pressure values ​​is the first-order difference of the pressure values ​​divided by the range of the pressure transmitter. The anchor point sequence is formed as follows: all effective paired sudden drop abrupt change points and corresponding gradual rise inflection points are arranged alternately in ascending order of sampling number. Each anchor point records its own sampling number, absolute timestamp and type label to form an anchor point sequence. Each feeding anchor point is followed by a slag discharge anchor point, and adjacent anchor point pairs are feeding followed by slag discharge or slag discharge followed by feeding.

2. The method for constructing a high-quality dataset for non-ferrous metal production processes according to claim 1, characterized in that, The output aligned multi-source data segment includes: Take two adjacent anchor points in the anchor point sequence in chronological order, and use the absolute timestamp of the previous anchor point as the start point of the interval and the absolute timestamp of the next anchor point as the end point of the interval to form an independent interval; Linearly map the original sampling timestamps of all data sources within each independent interval to normalized pseudo-time coordinates, so that the start point of the interval corresponds to the pseudo-time zero point and the end point of the interval corresponds to the pseudo-time one point, and save the mapping relationship between each original timestamp and pseudo-time. For each data source, points are taken on the pseudo-time axis at fixed steps, and linear interpolation is performed using the original sampled values ​​and the mapping relationship to obtain aligned data with equal pseudo-time intervals. All independent intervals are concatenated in chronological order to obtain aligned multi-source data segments, while saving the start and end absolute timestamps and interval numbers of each independent interval.

3. The method for constructing a high-quality dataset for non-ferrous metal production processes according to claim 2, characterized in that, The construction of the liveness markers includes: For each independent interval, extract the vibration sensor signal and the current sensor signal, and calculate the rate of change per unit time of adjacent sampling points; When the vibration change rate is greater than the preset vibration jump threshold and the current change rate is less than or equal to the current jump threshold, or when the current change rate is greater than the current jump threshold and the vibration change rate is less than or equal to the vibration jump threshold, the corresponding sampling point is marked as a suspected sensor failure, and the original signal value of the sampling point is retained. When both the vibration rate of change and the current rate of change exceed their respective thresholds and the direction of vibration change is the same as the direction of current change, the corresponding sampling point is marked as a real material event. For each original sampling point, the number of the independent interval to which it belongs is determined according to the original timestamp of the original sampling point. The corresponding pseudo-time value is found through the mapping relationship of the independent interval. The label of the original sampling point is assigned to the row with the smallest absolute value of the difference between the pseudo-time value and the pseudo-time value corresponding to the original sampling point in the same independent interval of the aligned multi-source data segment.

4. The method for constructing a high-quality dataset for non-ferrous metal production processes according to claim 3, characterized in that, The composite tracking code includes: During sampling, the sampling point number is read, the QR code of the sampling container is scanned to obtain the container number, the current process number is entered by the operator, and the two are combined into a composite tracking code. Write the composite tracking code into the sampling record table, and record the absolute sampling time at the same time; The composite tracking code is affixed to the sampling container in the form of a QR code. During testing, scanning the QR code will automatically associate the test results with the composite tracking code. Based on the composite tracking code, the sampling absolute time is retrieved, and an independent interval containing the sampling absolute time is found. If the sampling absolute time is exactly the common boundary of adjacent independent intervals, the previous independent interval is selected. The pseudo-time value corresponding to the sampling absolute time is calculated using the start and end absolute timestamps of the independent interval. In the data segment with the survival marker, the row with the smallest absolute value of the difference between the pseudo-time and the pseudo-time value within the same independent interval is selected, and the test result is written into the test result field of the row.

5. The method for constructing a high-quality dataset for non-ferrous metal production processes according to claim 4, characterized in that, When the absolute sampling time is exactly the common boundary of adjacent independent intervals, the previous independent interval is selected, and the mapping from the original sampling timestamp to the pseudo-time coordinate within the independent interval adopts a linear mapping relationship.

6. A system for constructing high-quality datasets for non-ferrous metal production processes, based on the method for constructing high-quality datasets for non-ferrous metal production processes as described in any one of claims 1 to 5, characterized in that: Also includes: The anchor point extraction module extracts natural physical anchor points from furnace temperature and pressure signals to obtain an anchor point sequence. The timing alignment module replaces the absolute clock with the anchor point sequence, matches anchor points across data sources to realign the time axis, and outputs the aligned multi-source data segments. The survival marker module adds a survival marker to each data segment, consisting of the synchronous change status of parallel vibration signals and current signals within the same window. A single signal jump marks the sensor as suspected failure and retains the original signal value of the sampling point, while multiple signal jumps simultaneously mark it as a real material event. The tracking and backlinking module generates a composite tracking code for each batch of sampled materials, consisting of the sampling location, container number, and operation sequence. This composite tracking code is carried throughout the entire process from sampling to testing, and the test results are forcibly backlinked to the production parameters corresponding to the sampling time in the data segment.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for constructing a high-quality dataset for non-ferrous metal production processes as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for constructing a high-quality dataset for non-ferrous metal production processes as described in any one of claims 1 to 5.

Images