Material database construction method and system based on high-throughput experiment multi-modal data

Abstract

The invention provides a material database construction method and system based on high-throughput experiment multi-modal data, and relates to the technical field of material database construction, and the method comprises the steps: building a standardized experiment data collection protocol; synchronously acquiring an original multi-modal experimental data set according to a standardized experimental data acquisition protocol; processing the infrared thermal imaging data, and extracting structured temperature characteristics of each catalyst channel; integrating the experimental metadata and the sensor time sequence data to generate a standardized sample-experiment-performance associated data record; and constructing a material database by adopting a three-layer logic architecture and a mixed physical storage strategy according to a standardized sample-experiment-performance associated data record and index information of an original multi-modal experiment data set. According to the method, the material database which is standard in data, accurate in feature, clear in structure and easy to manage can be constructed. The database can effectively support data retrieval, performance comparison and model construction in material research and development.

Description

Technical Field

[0001] The present invention relates to the technical field of material database construction, and in particular to a material database construction method and system based on high-throughput experimental multimodal data. Background Art

[0002] As a core component of materials genetic engineering, materials databases aim to systematically collect, store, manage, and share multi-dimensional information on materials, including composition, structure, process, and performance. By building a comprehensive materials database and combining it with technologies such as data mining and machine learning, the research and development of new materials can be accelerated, material properties can be optimized, and intelligent material design and screening can be achieved. Traditional materials database construction often relies on literature collection, manual data entry, and the integration of various experimental data. Its core lies in establishing an effective data model to describe material information and support efficient data retrieval and analysis, so that researchers can quickly locate and compare the characteristics of different materials.

[0003] With the widespread application of high-throughput experimental techniques in new materials fields such as catalysis, energy, and biology, the rate and volume of experimental data generated have increased dramatically, exhibiting typical multimodal characteristics. This data is not only high-dimensional and highly heterogeneous, but also requires extremely high temporal synchronization, spatial correspondence, and data integrity verification to ensure that subtle changes in materials during dynamic reactions and their correlation with performance can be accurately captured. Current methods for constructing materials databases cannot meet these requirements.

[0004] Patent application CN116049160A, filed in China, discloses a method for constructing a machine learning database based on multi-series aluminum alloy data. This method, specifically targeting the aluminum alloy field, addresses the data concentration and scarcity of specific element data in single-series modeling by integrating data from multiple heat-treatable aluminum alloy series, including Al-Zn-Mg-Cu, Al-Cu, and Al-Li. The database construction relies primarily on obtaining data on the alloy composition, processing technology (such as heat treatment schedules), and corresponding material properties of aluminum alloys from existing published literature. This method incorporates the unique characteristics of the aluminum alloy field, namely the complex relationship between multi-series alloy properties and composition and processing. It also employs data cleaning and boundary learning using machine learning algorithms (such as support vector regression) to fill in gaps in the data. The method attempts to construct a three-dimensional dataset encompassing alloy composition, heat treatment schedules, and material properties. However, the technical means of constructing the database by this method mainly rely on secondary processing and machine learning filling of existing literature data, which may lead to deficiencies in the data's precision of the original experimental process, the synchronization of multi-source heterogeneous data (especially dynamic process data), and deep mining of inherent correlations; at the same time, its data acquisition method also limits the direct integration and utilization of real-time, multimodal raw data (such as in-situ characterization data and sensor network time series data) during the experiment, and the accuracy and reliability of the data filled by machine learning are also highly dependent on the quality and coverage of the initial literature data. If the original literature data itself lacks a comprehensive record of complex experimental details, it may affect the integrity of the final database and the accuracy of the prediction model. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for constructing a materials database for the large amount of complex and heterogeneous multimodal data generated by high-throughput experiments, aiming to overcome the shortcomings of the existing technology in terms of data acquisition standardization, multi-source data synchronization and integration, key feature extraction accuracy, and database structuring and management efficiency, so as to efficiently and systematically construct a comprehensive, highly available materials database that includes sample information, experimental conditions, process data, core performance characteristics and raw data indexes.

[0006] The technical solution of the present invention is achieved as follows: In one aspect, the present invention provides a method for constructing a materials database based on high-throughput experimental multimodal data, comprising: S1. Establish a standardized experimental data acquisition protocol, including: using a sample coding system to assign unique identifiers to samples and record sample chemical composition, preparation method, and channel location information; using a multi-device time synchronization mechanism; and setting the adaptive acquisition frequency parameters of the infrared thermal imager; S2. Synchronously collect and record experimental metadata, infrared thermal imaging data, and sensor time series data according to the standardized experimental data acquisition protocol, and perform integrity verification on the collected data to obtain the original multimodal experimental dataset with timestamp and verification; S3. Processing the infrared thermal imaging data in the original multimodal experimental data set, extracting the structured temperature features of each catalyst channel through image preprocessing, channel calibration, and feature calculation, wherein the structured temperature features include temporal temperature features and spatial temperature distribution features; S4. Integrate the experimental metadata and sensor time series data in the original multimodal experimental dataset, combine them with the structured temperature features, and generate standardized sample-experiment-performance correlation data records through time alignment and normalization. S5. Based on the standardized sample-experiment-performance association data records and the index information of the original multimodal experimental dataset, a three-tier logical architecture and hybrid physical storage strategy are used to construct a material database.

[0007] Preferably, the sample coding system includes assigning a globally unique sample identifier to each catalyst sample, and recording the chemical composition, preparation method and channel position coordinate information of the sample in the screening device; the multi-device time synchronization mechanism includes using the network time protocol to perform millisecond-level clock synchronization on the infrared thermal imager, temperature sensor, mass flow controller, pressure sensor and central control system before the experiment; the adaptive acquisition frequency parameters of the infrared thermal imager include reducing the acquisition frequency during the stable reaction stage and increasing the acquisition frequency during the temperature mutation stage.

[0008] Preferably, the synchronously collected and recorded experimental metadata include: unique experimental identifier, experimental date, experimenter information, experimental device identification, sample batch information, experimental batch information, temperature program setting value, target flow rate and reaction pressure of each gas component; the infrared thermal imaging data includes the original temperature matrix data recording the temperature change of the catalyst channel; the sensor timing data includes the platform thermocouple temperature value, gas flow value and pressure value.

[0009] Preferably, the image preprocessing includes geometric correction of the infrared image sequence; the channel calibration includes using a pre-stored standard channel template image of the screening device to align each frame of the infrared image with the template through image registration, and accurately segmenting the region of interest of each channel; the feature calculation includes extracting the time series and spatial distribution features of the temperature data within the region of interest of each channel.

[0010] Preferably, the temporal temperature characteristics include ignition time, ignition temperature, maximum temperature rise, peak temperature and its arrival time, average temperature in the reaction stable stage and standard deviation of temperature fluctuation; the spatial temperature distribution characteristics include the temperature standard deviation and hot spot area ratio in the region of interest.

[0011] Preferably, the ignition time in the time series temperature characteristic is calculated by normalizing the dynamic ignition criterion To determine, when The ignition time is determined when the preset ignition criterion threshold value is continuously exceeded for at least a preset time period, wherein Calculated by the following formula: , in, is the average channel temperature at time i, is the average channel temperature at the previous moment, is the time interval between time i and time i-1, is the normalized temperature rise rate parameter, is the reference baseline temperature, is the characteristic activation temperature parameter, is the index adjustment factor.

[0012] Preferably, the time alignment includes accurately aligning the infrared thermal imaging data with the sensor time series data on the time axis based on a high-precision timestamp; the standardization processing includes converting data from different sources into a unified international system of units, eliminating sensor failure noise points, extracting statistics of time-varying parameters in key reaction stages, and forming standardized records.

[0013] Preferably, the three-layer logical architecture includes a basic metadata layer, a derived feature data layer and a raw data index layer; the hybrid physical storage strategy includes storing structured metadata and feature data through a relational database, and storing large-capacity raw infrared thermal imaging video files and sensor log files through a distributed object storage system.

[0014] Preferably, the basic metadata layer includes a sample information table, a preparation method table, an experimental configuration table and a channel mapping table; the derived feature data layer includes a catalyst performance table and a time series data summary table; the raw data index layer includes a raw data registry and a key frame table; and a standardized application interface is provided to support data operations and integration.

[0015] In another method, the present invention also provides a material database construction system based on high-throughput experimental multimodal data, wherein the system is used to implement any of the methods described above, and the system comprises: The data protocol management module is used to establish a standardized experimental data acquisition protocol. It includes a sample encoding submodule for assigning unique identifiers to samples and recording their chemical composition, preparation method, and channel location information; a time synchronization submodule for achieving millisecond-level clock synchronization of experimental equipment; and an acquisition frequency control submodule for setting the adaptive acquisition frequency parameters of the infrared thermal imager. The multimodal data acquisition module is used to synchronously collect and record experimental metadata, infrared thermal imaging data, and sensor time series data according to the standardized experimental data acquisition protocol, and perform data integrity verification to generate a timestamped and verified original multimodal experimental dataset; The infrared image processing module is used to process the infrared thermal imaging data in the original multimodal experimental dataset. It includes an image preprocessing unit for geometric correction of the infrared image sequence; a channel calibration unit for accurately aligning the infrared image with the standard channel template and segmenting the channel area; and a feature calculation unit for extracting structured temperature features including temporal temperature features and spatial temperature distribution features. The data integration and association module is used to integrate the experimental metadata and sensor time series data in the original multimodal experimental dataset, and combine them with structured temperature features to generate standardized sample-experiment-performance association data records through time alignment and normalization. The database management module is used to build and maintain the material database based on standardized sample-experiment-performance association data records and index information of the original multimodal experimental data set, using a three-tier logical architecture and a hybrid physical storage strategy. The three-tier logical architecture includes a basic metadata layer, a derived feature data layer, and a raw data index layer.

[0016] The present invention has the following beneficial effects compared to the prior art: (1) By systematically collecting, processing, integrating, and storing multimodal data generated by high-throughput experiments, the present invention can construct a materials database with standardized data, accurate features, clear structure, and ease of management. This database can effectively support data retrieval, performance comparison, and model building in materials research and development, thereby improving the efficiency of new material screening and the reliability of performance prediction; (2) The present invention implements a standardized data acquisition protocol during the data acquisition phase, including assigning a globally unique sample identifier to each catalyst sample and recording its channel location information, using the network time protocol to synchronize the clocks of multiple devices at the millisecond level, and dynamically adjusting the acquisition frequency of the infrared thermal imager according to the reaction stage. This ensures the accuracy, spatiotemporal correspondence, and integrity of the original data from the source. (3) The present invention uses image preprocessing and calibration techniques, including geometric correction, precise channel calibration based on pre-stored standard template images, and region of interest segmentation, for the processing of infrared thermal imaging data. On this basis, the temporal and spatial distribution characteristics are calculated, thereby improving the accuracy and reliability of performance indicators obtained from complex infrared data. (4) This invention precisely aligns infrared thermal imaging data with sensor time series data on the time axis based on high-precision timestamps, converts data from different sources into a unified international system of units, eliminates sensor failure noise points, and extracts statistics of time-varying parameters in key reaction stages, thus forming a standardized sample-experiment-performance correlation record. This systematic data integration and standardization process effectively solves the problem of inconsistencies in format, unit, and time base among multi-source heterogeneous data, ensuring the correct correlation and comparability between data of different dimensions; (5) The present invention uses a three-tiered logical architecture and a hybrid physical storage strategy combining a relational database with distributed object storage to construct a materials database. Key performance parameters, such as ignition time, determined based on normalized dynamic ignition criteria, are explicitly stored in the catalyst performance table of the derived feature data layer. This specific database architecture design and storage method not only enables efficient management and storage of structured metadata, derived feature data, and large-capacity raw data files, ensuring data integrity and accessibility, but also optimizes data retrieval efficiency through clear logical stratification and direct storage of key features, facilitating rapid access to required information and in-depth analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 is a flow chart of the method of the present invention; Figure 2 It is a technical implementation diagram of the present invention; Figure 3 This is a data collection schematic diagram of the present invention; Figure 4 Schematic diagram of the feature extraction process of the present invention; Figure 5 A schematic diagram of a data record generation process of the present invention; Figure 6 Schematic diagram of the system module of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] like Figure 1 As shown, the present invention provides a method for constructing a material database based on high-throughput experimental multimodal data, comprising: S1. Establish a standardized experimental data acquisition protocol, including: using a sample coding system to assign unique identifiers to samples and record sample chemical composition, preparation method, and channel location information; using a multi-device time synchronization mechanism; and setting the adaptive acquisition frequency parameters of the infrared thermal imager; S2. Synchronously collect and record experimental metadata, infrared thermal imaging data, and sensor time series data according to the standardized experimental data acquisition protocol, and perform integrity verification on the collected data to obtain the original multimodal experimental dataset with timestamp and verification; S3. Processing the infrared thermal imaging data in the original multimodal experimental data set, extracting the structured temperature features of each catalyst channel through image preprocessing, channel calibration, and feature calculation, wherein the structured temperature features include temporal temperature features and spatial temperature distribution features; S4. Integrate the experimental metadata and sensor time series data in the original multimodal experimental dataset, combine them with the structured temperature features, and generate standardized sample-experiment-performance correlation data records through time alignment and normalization. S5. Based on the standardized sample-experiment-performance association data records and the index information of the original multimodal experimental dataset, a three-tier logical architecture and hybrid physical storage strategy are used to construct a material database.

[0021] like Figure 2As shown, the technical implementation process of the present invention includes: 1) establishment and execution of a standardized data acquisition protocol, namely, first setting a globally unique sample identifier for each catalyst sample and recording its composition, preparation method, and precise channel position in the screening device. Before the experiment, the infrared thermal imager, various sensors, and central control system are synchronized with the clock at the millisecond level using the network time protocol. At the same time, the infrared thermal imager's adaptive acquisition frequency is set so that it can reduce the frequency during the stable reaction phase and increase the frequency during the rapid temperature change phase; 2) synchronous acquisition and preliminary verification of multimodal experimental data, namely, synchronously acquiring and recording detailed experimental metadata, infrared thermal imaging raw temperature matrix sequence, and sensor timing data during the experiment, and performing a preliminary integrity check after acquisition; 3) Deep processing and key feature extraction of infrared thermal imaging data, namely, geometric correction of the infrared image sequence, using pre-stored standard channel template images to accurately segment the region of interest of the channel where each catalyst sample is located through image registration, and then calculating its temporal temperature characteristics for each channel, especially the ignition time. The ignition time is determined by calculating the normalized dynamic ignition criterion. This criterion comprehensively considers the temperature rise rate and the normalization degree of the current temperature relative to the reference baseline and the characteristic activation temperature and introduces an exponential adjustment factor. When this criterion continuously exceeds the preset threshold for a preset time, it is determined as the ignition moment. In addition, other temporal features such as ignition temperature, maximum temperature rise, peak temperature and spatial temperature distribution characteristics such as hot spot area ratio are extracted; 4) Integration, standardization, and formation of associated records of multi-source data, namely, integrating the extracted structured temperature features with experimental metadata and sensor time series data, accurately aligning infrared thermal imaging data and sensor time series data on the time axis based on high-precision timestamps, performing unit unification, noise removal, and extraction of time-varying parameter statistics of key reaction stages, ultimately forming standardized, structured sample-experiment-performance associated data records containing sample information, experimental conditions, processed sensor data, and key performance characteristics; 5) Construction and management of the materials database, namely, adopting a three-layer logical architecture consisting of a basic metadata layer, a derived feature data layer, and a raw data index layer. For physical storage, a relational database is used to store structured metadata and derived feature data, and a distributed object storage system is used to store large-capacity raw infrared video and sensor logs. Key performance parameters such as ignition time determined by normalized dynamic ignition criteria are stored in the catalyst performance table of the derived feature data layer. Finally, a standardized application program interface is provided to support database data operations and application analysis.

[0022] Specifically, if Figure 3 As shown, in one embodiment of the present invention, step S1 includes: Constructing a sample coding system involves assigning a globally unique sample identifier to each catalyst sample and recording the sample's chemical composition, preparation method, and channel position coordinate information within the screening device. This sample identifier is designed as a structured code, integrating, for example, the project code, experimental batch number, material classification code, and the sample's serial number within that batch. It can also be combined with information about the sample's physical location within the screening device, such as the reaction plate number and well row and column numbers. This structured identifier not only ensures the absolute uniqueness of each sample but also embeds metadata within the identifier itself, allowing for easy parsing and preliminary screening. Simultaneously recorded with the sample identifier is the sample's detailed chemical composition information, accurate to the percentage of each component, purity, and raw material source, as well as the sample's exact three-dimensional channel position coordinate information within the high-throughput screening device.

[0023] A multi-device time synchronization mechanism is employed, including millisecond-level clock synchronization of infrared thermal imagers, temperature sensors, mass flow controllers, pressure sensors, and the central control system using the Network Time Protocol (NTP) before the experiment. Specifically, before each round of high-throughput experiments officially begins, the central control system deployed on the experimental platform will perform mandatory clock synchronization calibration on all devices involved in data acquisition using a network time protocol (NTP) or, in scenarios with higher time accuracy requirements, the Precision Time Protocol (PTP / IEEE1588). These devices include infrared thermal imagers for capturing the surface temperature distribution of the sample, thermocouples or resistance temperature sensors for monitoring specific point temperatures in each channel, mass flow controllers for precisely controlling the feed of reaction gases, pressure sensors for monitoring the ambient pressure of the reaction system, and the central data recording and control system itself, which is responsible for aggregating and initially processing data. The goal is to keep the timestamp error of all devices to the millisecond level or even lower, thereby providing a unified and reliable time reference for subsequent fusion analysis of multimodal data.

[0024] As an implementation method, a hardware synchronization pulse signal is designed to be uniformly issued by the central control system when the experiment starts or a specific trigger event occurs. This signal is distributed in parallel to all key data acquisition devices and serves as a common zero-time reference point or periodic calibration signal to further improve the time synchronization accuracy of data recording across devices and effectively eliminate slight time deviations that may be introduced by factors such as network delay jitter.

[0025] The infrared thermal imager's adaptive acquisition frequency parameters are set, including lowering the acquisition frequency during stable reaction phases and increasing it during temperature fluctuations. This setting aims to strike a balance between fully capturing key dynamic characteristics of the reaction process and effectively controlling the total amount of data stored. Specifically, when the infrared thermal imager determines, through real-time image analysis or associated sensor data, that the reaction process for all sample channels is relatively stable, meaning the temperature change rate is low or within a preset stability threshold, the system automatically adjusts the infrared thermal imager's acquisition frequency to a lower baseline level, such as one frame per second or lower, to avoid generating a large amount of redundant thermal imaging data during the stable reaction period. However, if the system detects a dramatic temperature change in any one or more sample channels, or if the pre-set experimental procedure predicts the onset of a temperature fluctuation, such as catalyst ignition, flameout, or significant heat release or endothermic events, the infrared thermal imager's acquisition frequency is quickly and automatically increased to a preset higher level, such as tens or even higher frames per second, to ensure that the dynamic evolution and subtle characteristics of these key reaction events are accurately recorded with sufficient temporal resolution.

[0026] In this embodiment, the acquisition frequency of the infrared thermal imager is Designed based on real-time monitoring of the average temperature change rate of the sample area The function is dynamically adjusted as follows: , The calculation result Will be constrained to the preset minimum acquisition frequency and the maximum acquisition frequency between. Represents the minimum data collection frequency to ensure the stable reaction period. Represents the highest acquisition frequency allowed by the instrument or data processing capability. It is an adjustable sensitivity coefficient used to amplify or reduce the impact of the temperature change rate on the acquisition frequency. It refers to the temperature change rate obtained by real-time differential calculation of the temperature data of a specific area of ​​interest of the infrared thermal imager, and abs represents the absolute value of the temperature change rate.

[0027] Specifically, in one embodiment of the present invention, in step S2, the synchronously collected and recorded experimental metadata includes: the unique experimental identifier, the experimental date, the experimenter information, the experimental device identification, the sample batch information, the experimental batch information, the temperature program setting value, the target flow rate of each gas component and the reaction pressure; the infrared thermal imaging data includes the original temperature matrix data recording the temperature changes of the catalyst channel; the sensor timing data includes the platform thermocouple temperature value, the gas flow value and the pressure value.

[0028] Specifically, after metadata is entered, it is stored in a structured file format such as JSON or XML. After storage, a hash value is generated for each set of metadata records.

[0029] In this embodiment, the infrared thermal imager continuously captures infrared radiation from the sample area at an adaptive acquisition frequency during the experiment and converts it into a two-dimensional temperature matrix sequence, i.e., a raw thermal imaging video stream. The captured data is recorded in a raw binary format with detailed timestamps or in a standard image sequence format.

[0030] In this embodiment, sensor time series data is collected at a preset higher fixed frequency or a configurable adaptive frequency. Each data point is accompanied by a timestamp, and the data is stored in CSV format.

[0031] In this embodiment, after the initial completion of the data collection task, a data verification process is also included to quickly assess the quality of the collected data. Verification includes checking whether all expected experimental metadata fields are correctly populated and their values ​​are within a reasonable range; verifying whether the infrared thermal imaging data file is complete, whether the file size meets the expected size, and whether it can be correctly parsed; confirming the existence of all sensor time series data streams, whether the timestamps are continuous and aligned with the start and end times of the experiment, and whether the sensor readings are within the preset physical meaning or safety threshold range.

[0032] Specifically, set the data health score To quantify the initial quality of the data. If the verification detects any missing, corrupted, or significantly abnormal data, such as a data health score below a preset threshold, the system automatically generates an alert and marks the relevant data record, prompting the experimenter to manually review or decide whether to repeat part of the experiment or data collection.

[0033] In a specific example, data health scoring The calculation process is as follows: , Where, 、 、 Represent the quality scores of metadata, infrared data and sensor data respectively, 、 、 are the corresponding weight coefficients respectively.

[0034] , Where, Indicates the total number of metadata fields. Indicates the number of fields that have been successfully filled. is a hyperparameter that controls the degree of penalty for missing fields. is the quality score of the ith field, ranging from 0 to 1, and is assessed by verifying whether the field value is within the expected range, is in the correct format, and is logically consistent with other related fields. 、 They are the sensitivity parameter and threshold parameter of quality assessment respectively.

[0035] , Where, is the number of infrared image frames actually acquired, The expected number of frames to be obtained based on the experiment duration and the set acquisition frequency is: is an exponential parameter that adjusts the integrity weight. The second term evaluates the continuity of infrared data, where is the total number of time sampling points, Indicates the key data features at time t, such as the average temperature value, is a parameter that controls the importance of continuity. The third term evaluates the signal-to-noise ratio (SNR), 、 are the sensitivity parameter and reference threshold for signal-to-noise ratio evaluation, respectively.

[0036] , Where K is the total number of sensors, is the number of outliers for the kth sensor, is the total number of data points for this sensor, The second term evaluates the collaborative consistency between sensors. is the mean absolute deviation of the correlation coefficients between all sensor pairs, 、 are the reference threshold and sensitivity parameters for controlling collaborative evaluation. The third term introduces the KL divergence in information theory. Measures the difference between the sensor data distribution P and the expected theoretical distribution Q, Controls the weight of this item.

[0037] In the calculation Then, compare it with the preset threshold Compare, if , the alarm mechanism is triggered. The alarm level is determined based on the following formula: , The alarm level is quantified from 1 to 4 in this way, with level 4 being the most serious and requiring immediate intervention by the experimenter and possibly the need to re-execute the experiment, while level 1 is only a warning. Represents the ceiling function.

[0038] Specifically, if Figure 4 As shown, in one embodiment of the present invention, in step S3, the processing of infrared thermal imaging data includes image preprocessing, channel calibration and feature calculation. The image preprocessing includes geometric correction of the infrared image sequence; the channel calibration includes using the pre-stored standard channel template image of the screening device to align each frame of the infrared image with the template through image registration, and accurately segmenting the region of interest of each channel; the feature calculation includes extracting the time series and spatial distribution features of the temperature data within the region of interest of each channel. The time series temperature features include ignition time, ignition temperature, maximum temperature rise, peak temperature and its arrival time, average temperature in the reaction stable stage and temperature fluctuation standard deviation; the spatial temperature distribution features include temperature standard deviation and hot spot area ratio within the region of interest.

[0039] In a specific example, the implementation process of step S3 is as follows: S3.1 Image Preprocessing Geometric correction is applied to the captured infrared image sequence. This process uses distortion parameters previously acquired by photographing a standard calibration target to establish and apply a polynomial transformation model to inversely map the pixel coordinates of each frame. This step corrects image distortion introduced by factors such as optical distortion of the infrared camera lens, non-perpendicular mounting angles, and sample stage tilt, ensuring the correct spatial position and geometric form of objects in the image. Furthermore, a median filter is applied to smooth the corrected image to suppress sensor noise while preserving image edges and details, thereby improving the image signal-to-noise ratio.

[0040] S3.2 Channel Calibration Based on the geometrically corrected infrared images, precise channel demarcation and automatic segmentation of the regions of interest (ROIs) for each channel are performed. This process utilizes image registration based on the scale-invariant feature transform (SIFT) algorithm to precisely spatially align each live infrared image frame with a pre-stored template image of a standard channel corresponding to a high-throughput screening device. This template image is precisely pre-created and contains the idealized geometric outline, center position, and unique logical number of each catalyst sample channel. Through image registration, a precise affine or perspective transformation matrix is ​​determined. This transformation matrix is ​​used to accurately map the pre-defined channel region masks or coordinate lists on the template image onto the live infrared image, thereby automatically and accurately segmenting the actual regions of interest occupied by each catalyst sample. As an implementation approach, a deep learning instance segmentation model based on the Mask R-CNN architecture, pre-trained through supervised learning on a large dataset of infrared images with manually annotated channel positions and outlines, is employed to directly identify and segment the precise boundaries of each sample channel from the input infrared image. This method offers improved segmentation accuracy and robustness under conditions of closely spaced channels, partial occlusion, or complex backgrounds.

[0041] S3.3 Feature calculation For each precisely segmented sample channel ROI, its temporal temperature characteristics and spatial temperature distribution characteristics are calculated. The ignition time is determined by calculating the following normalized dynamic ignition criterion (NDIC): , in, is the normalized dynamic ignition criterion value calculated at time i, is the average temperature in the sample channel ROI at time i, is the average channel temperature at the previous moment, is the time interval between time i and time i-1, that is, the sampling period of the infrared thermal imager, is the normalized temperature rise rate parameter, As the reference baseline temperature, the average value of the ROI temperature during a stable period after the reaction starts is taken. is the characteristic activation temperature parameter, which is the typical temperature threshold at which the catalyst is significantly activated or undergoes a violent reaction, set based on the prior knowledge of the catalyst system. is an exponential adjustment factor, set to a constant greater than 1, used to amplify the criterion response when the temperature approaches or exceeds the characteristic activation temperature.

[0042] The default minimum confirmation points are , for example, the number of sampling frames corresponding to 0.5 seconds. Continuous values The sampling points all stably exceed the preset judgment threshold , the first one that meets this condition will be The time point corresponding to the value The NDIC criterion combines the temperature rise rate and the normalization degree of the current temperature relative to the reference baseline and the characteristic activation temperature, and introduces an exponential adjustment factor to robustly and adaptively identify the ignition moment.

[0043] In addition to the ignition time, other temporal features extracted include: ignition temperature (i.e., the average temperature within the ROI at the determined ignition time point), maximum temperature rise (the difference between the peak temperature recorded within the ROI during the experiment and the initial baseline temperature), peak temperature (the highest temperature value recorded within the ROI during the experiment), and its reaching time.

[0044] Extracting spatial temperature distribution features involves calculating the temperature standard deviation and hot spot area ratio within the ROI. The temperature standard deviation within the ROI characterizes the uniformity of the temperature distribution within the ROI. To further quantify the non-uniformity of the spatial temperature distribution, the hot spot area ratio is calculated: a dynamic high temperature threshold is set (for example, the average temperature within the ROI at the current time point plus twice its standard deviation). All pixel areas within the ROI with temperatures above this threshold are identified as hot spots. The ratio of the total area of ​​these hot spots to the total area of ​​the ROI is the hot spot area ratio. This ratio reflects the proportion of high-temperature areas on the sample surface.

[0045] Specifically, if Figure 5 As shown, in one embodiment of the present invention, step S4 includes: The sensor time series data collected in step S2, such as platform thermocouple temperature, gas flow, and pressure values, are precisely matched and aligned with the various time series temperature features extracted from the infrared thermal imaging data in step S3, such as ignition time, ignition temperature, maximum temperature rise, peak temperature, and its arrival time, on a unified time axis. To address the potential sampling frequency inconsistencies between different data sources, an interpolation algorithm is employed. For sensor data with relatively low sampling frequencies, Lagrange interpolation or cubic spline interpolation is applied at key time points, such as before and after catalyst ignition, to generate synchronized data points that precisely correspond in time to the high-frequency infrared data. As an implementation method, a dynamic time warping (DTW) algorithm based on cross-correlation analysis is introduced. This algorithm can more flexibly handle potential nonlinear time delays and deformations between different sensor signals, thereby achieving more robust and accurate time alignment.

[0046] Data from different sources are standardized to ensure that all physical quantities are expressed in SI or industry-standard units. For example, temperatures are uniformly converted to Kelvin, gas flows are standardized to standard milliliters per minute, and pressures are standardized to kilopascals. Secondly, to ensure data quality, the system implements a noise data removal process using statistically based methods, specifically the Local Outlier Factor (LOF) algorithm, to identify and mark significant anomalous data points caused by sensor failures or strong external interference. Subsequently, core statistics are calculated for parameters that vary over time during the experiment, such as actual gas flow, reactor pressure, and programmed substrate temperature, at key reaction stages, such as reactant introduction, programmed temperature ramp, catalyst ignition, reaction stabilization, and cooling. These statistics, including mean, median, standard deviation, integral, maximum, minimum, and slope of the linear fit, are incorporated into standardized data records as condensed representations of the original time series data.

[0047] After time alignment and standardization, a standardized sample-experiment-performance correlation data record is generated for each catalyst sample. This record is a structured data set that logically and clearly links sample information, the experimental conditions it actually experienced, and the catalytic performance it exhibited. Specifically, it includes: the sample's unique identifier, chemical composition, preparation method, and channel position coordinates in the screening device; the experiment's unique identifier, experimental date, experimenter information, experimental device identification, sample batch information, experimental batch information, detailed temperature program settings, target flow rates for each gas component, and total reaction pressure; as well as all temporal temperature features extracted from the infrared thermal imaging data (such as ignition time, ignition temperature, maximum temperature rise, peak temperature and its arrival time, average temperature during the reaction stabilization phase, and standard deviation of temperature fluctuations) and spatial temperature distribution features (such as temperature standard deviation and hot spot area ratio within the region of interest).

[0048] Specifically, in one embodiment of the present invention, step S5 includes: First, in the design of the three-tier logical architecture, the database is divided into the basic metadata layer, the derived feature data layer, and the original data index layer. Each layer has a clear functional positioning and data content.

[0049] The basic metadata layer includes: a sample information table, which is used to record the globally unique identifier of each catalyst sample, detailed chemical composition, specific preparation method or process parameters, and its precise channel position coordinate information in the screening device; a preparation method table, which can serve as a detailed supplement to the preparation method in the sample information table, and provides a structured description of different preparation processes, parameters, and batches of raw materials used; an experiment configuration table, which is used to store the detailed setting parameters of each high-throughput experiment, such as the experiment unique identifier, the experiment execution date, operator information, the unique identifier of the experimental device used, sample batch information, experiment batch information, preset temperature program curve parameters, the target flow rate and concentration of each gas component, and the total reaction pressure setting value of the system; a channel mapping table, which is used to accurately define the correspondence between physical channels and logical samples in the screening device.

[0050] The derived feature data layer stores data extracted and processed from the raw data that directly reflects the key characteristics of the material properties and experimental process. This layer includes: a catalyst performance table, which is linked to the sample information table and the experimental configuration table via foreign keys and stores the various catalytic performance indicators exhibited by each sample under specific experimental conditions. These indicators are mainly derived from the structured temperature features extracted in step S3, such as ignition time, ignition temperature, maximum temperature rise, peak temperature and its arrival time, average temperature and temperature fluctuation standard deviation in the stable reaction phase, temperature standard deviation in the region of interest, and hot spot area ratio; a time series data summary table, which is used to store statistics at key reaction stages obtained after standardizing the sensor time series data in step S4, such as the actual average flow rate of each gas component, the actual average pressure in the reactor, and the average temperature recorded by the platform thermocouple.

[0051] The raw data index layer is responsible for managing and linking to massive amounts of raw experimental data. This layer includes: a raw data registry, which assigns a unique internal identifier to each raw data file (such as infrared thermal imaging video and sensor log files) and records metadata such as its storage path (which may point to a location in a distributed object storage system), file type, size, creation time, and associated experiment and sample IDs; and a keyframe table. For infrared thermal imaging video data, this table records the time frame index or precise timestamp corresponding to key experimental events in the video (such as the moment of ignition, reaching peak temperature, and the start of reaction stabilization), as well as the offset or direct access link of these keyframes in the raw video file. This allows for rapid location and extraction of raw image data at important time nodes without having to fully decode the entire video file.

[0052] In a specific example, a simple example of a three-tier logical architecture is as follows: First layer: basic metadata layer Table 1 Sample information table Field Name Data Type Constraints / Instructions Example SampleID text Primary key, sample unique identifier "CAT-20240512-001" Chemical Composition text Chemical composition description <![CDATA["Pt 0.5% / Al2O3"]]> PreparationMethodID text Foreign key, associated to the preparation method table "PM-SprayDry-003" SynthesisDate date Sample preparation date "2024-03-15" BatchNumber text Sample batch number "SN20240315-A" PhysicalForm text Sample physical form "powder" Supplier text Raw material suppliers "Sigma-Aldrich" Notes text Other notes "High specific surface area carrier" Table 2 Preparation method Field Name Data Type Constraints / Instructions Example PreparationMethodID text Primary key, unique identifier of the preparation method "PM-SprayDry-003" MethodName text Method Name "Spray drying method" MethodDescription text Detailed process step description "The precursor solution is atomized and dried in a high-temperature air flow..." KeyParameters JSON / text Key process parameters "{'inlet_temp': '180C', 'feed_rate': '5ml / min'}" Table 3 Experimental configuration table Field Name Data Type Constraints / Instructions Example ExperimentID text Primary key, unique identifier of the experiment "EXP-HTS-20240513-A-01" ExperimentDate date Experiment execution date "2024-05-13" OperatorName text Experiment operator name "Zhang San" DeviceID text Unique identification of the screening device used "HTS-Device-002" TemperatureProgram Text / JSON Preset temperature programs "Raise from 50C to 600C at a rate of 10C / min" Gas Composition Text / JSON Gas components and target flow rates "{'CH4': '50sccm', 'O2': '100sccm', 'N2': 'balance'}" TotalFlowRateSet Numerical Total gas target flow rate (sccm) 170 PressureSet Numerical System target reaction pressure (kPa) 101.3 ExperimentNotes text Experimental Notes "System pressure fluctuates slightly" Table 4 Channel mapping table Field Name Data Type Constraints / Instructions Example MappingID Integer primary key 1001 ExperimentID text Foreign key, associated to the experiment configuration table "EXP-HTS-20240513-A-01" ReactorChannelNo Integer Physical channel number in the screening device 5 SampleID text Foreign key, associated to the sample information table "CAT-20240512-001" LoadDate date Sample loading date "2024-05-12" Second layer: derived feature data layer Table 5 Catalyst performance Field Name Data Type Constraints / Instructions Example PerformanceID Integer primary key 2001 ExperimentID text Foreign key, associated to the experiment configuration table "EXP-HTS-20240513-A-01" SampleID text Foreign key, associated to the sample information table "CAT-20240512-001" IgnitionTime Numerical Ignition time (seconds) 125.5 IgnitionTemperature Numerical Ignition temperature (°C or K) 350.2 MaxTemperatureRise Numerical Maximum temperature rise (°C or K) 150.7 PeakTemperature Numerical Peak temperature (°C or K) 500.9 TimeAtPeakTemperature Numerical Time to reach peak temperature (seconds) 180.3 SteadyStateTemperature Numerical Average temperature during the stable reaction phase 480.5 ROITemperatureStdDev Numerical Standard deviation of temperature within the region of interest 5.3 HotspotAreaRatio Numerical Hot spot area ratio 0.15 Table 6 Time series data summary table Field Name Data Type Constraints / Instructions Example SummaryID Integer primary key 3001 ExperimentID text Foreign key, associated to the experiment configuration table "EXP-HTS-20240513-A-01" SampleID text Foreign key, associated to the sample information table "CAT-20240512-001" Stage text Experimental stage "Steady-state period" ActualAvgFlowRate Numerical Actual average flow rate (sccm) 49.8 ActualAvgPressure Numerical Actual average reaction pressure (kPa) 101.1 PlatformAvgTemperature Numerical Average temperature of platform thermocouples (°C) 485.2 DurationOfStage Numerical Phase duration (seconds) 600 The third layer: raw data index layer Table 7 Raw data registration table Field Name Data Type Constraints / Instructions Example RawDataID Integer Primary key, unique identifier of original data 4001 ExperimentID text Foreign key, associated to the experiment configuration table "EXP-HTS-20240513-A-01" SampleID text Foreign key, associated to the sample information table "CAT-20240512-001" DataType text Data Type "IRVideo" FileName text Original file name "EXP-HTS-20240513-A-01_Sample001_IR.mp4" StoragePath text The URI or path of the data in the distributed storage "s3: / / my-hts-data / videos / 2024 / 05 / ir_vid_001.mp4" FileSizeMB Numerical File size (MB) 1024.5 CreationTimestamp Timestamp File creation time "2024-05-13 15:30:00" Checksum text File checksum "a1b2c3d4e5f6..." Table 8 Key frame table Field Name Data Type Constraints / Instructions Example KeyframeID Integer Primary Key 5001 RawDataID Integer Foreign key, linking to the original data registry 4001 EventName text Key event name "IgnitionStart" EventTimestampVideo Numerical Timestamp of the event in the video (seconds or frame number) 125.5 (seconds) CorrespondingExperimentTime Numerical The time of the event in the entire experiment timeline (seconds) 305.5 Visual Features BLOB / Text Path to store extracted visual feature descriptors " / features / kf_5001_sift.bin" ThumbnailPath text The storage path of keyframe thumbnails " / thumbnails / kf_5001.jpg" As can be seen from the eight tables above, the basic metadata layer defines "what" and "under what conditions"; the derived feature data layer records "how it behaves"; and the raw data index layer provides "where the original evidence is." This hierarchical structure facilitates data management, querying, analysis, and system maintainability.

[0053] Specifically, structured metadata and derived feature data—that is, the content of the basic metadata layer and derived feature data layer—are stored in high-performance relational database management systems (RDBMSs) such as PostgreSQL or MySQL due to their relatively small size, neat structure, and frequent query and update operations. Large-capacity raw multimodal experimental datasets, particularly raw infrared thermal imaging video files and detailed sensor log files, are stored in lower-cost and more scalable distributed object storage systems (such as Amazon S3, MinIO, or Ceph) or network-attached storage (NAS) due to their large size and predominantly sequential or random large-block read access patterns. The storage path information in the raw data index layer points to specific objects in these distributed storage systems. This hybrid storage strategy effectively avoids the performance bottlenecks and management complexity associated with directly storing large binary files in relational databases.

[0054] Specifically, this embodiment also includes standardized application programming interfaces (APIs), such as RESTful APIs or GraphQL APIs. These APIs encapsulate CRUD (create, read, update, and delete) operations on the database's three-tier logical structure and support complex queries, data export, and seamless integration with other external systems (such as electronic laboratory notebooks (ELNs), laboratory information management systems (LIMSs), and data analysis platforms).

[0055] In addition, if Figure 6 As shown, the present invention also provides a material database construction system based on high-throughput experimental multimodal data, which is used to implement the method described above, characterized in that the system includes: The data protocol management module is used to establish a standardized experimental data acquisition protocol. It includes a sample encoding submodule for assigning unique identifiers to samples and recording their chemical composition, preparation method, and channel location information; a time synchronization submodule for achieving millisecond-level clock synchronization of experimental equipment; and an acquisition frequency control submodule for setting the adaptive acquisition frequency parameters of the infrared thermal imager. The multimodal data acquisition module is used to synchronously collect and record experimental metadata, infrared thermal imaging data, and sensor time series data according to the standardized experimental data acquisition protocol, and perform data integrity verification to generate a timestamped and verified original multimodal experimental dataset; The infrared image processing module is used to process the infrared thermal imaging data in the original multimodal experimental dataset. It includes an image preprocessing unit for geometric correction of the infrared image sequence; a channel calibration unit for accurately aligning the infrared image with the standard channel template and segmenting the channel area; and a feature calculation unit for extracting structured temperature features including temporal temperature features and spatial temperature distribution features. The data integration and association module is used to integrate the experimental metadata and sensor time series data in the original multimodal experimental dataset, and combine them with structured temperature features to generate standardized sample-experiment-performance association data records through time alignment and normalization. The database management module is used to build and maintain the material database based on standardized sample-experiment-performance association data records and index information of the original multimodal experimental data set, using a three-tier logical architecture and a hybrid physical storage strategy. The three-tier logical architecture includes a basic metadata layer, a derived feature data layer, and a raw data index layer.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for constructing a material database based on high-throughput experimental multimodal data, characterized in that: include: S1. Establish a standardized experimental data acquisition protocol, including: using a sample coding system to assign unique identifiers to samples and record sample chemical composition, preparation method, and channel location information; using a multi-device time synchronization mechanism; and setting the adaptive acquisition frequency parameters of the infrared thermal imager; S2. Synchronously collect and record experimental metadata, infrared thermal imaging data, and sensor time series data according to the standardized experimental data acquisition protocol, and perform integrity verification on the collected data to obtain the original multimodal experimental dataset with timestamp and verification; S3. Processing the infrared thermal imaging data in the original multimodal experimental data set, extracting the structured temperature features of each catalyst channel through image preprocessing, channel calibration, and feature calculation, wherein the structured temperature features include temporal temperature features and spatial temperature distribution features; S4. Integrate the experimental metadata and sensor time series data in the original multimodal experimental dataset, combine them with the structured temperature features, and generate standardized sample-experiment-performance correlation data records through time alignment and normalization. S5. Based on the standardized sample-experiment-performance association data records and the index information of the original multimodal experimental data set, a three-tier logical architecture and hybrid physical storage strategy are used to construct a material database.

2. The method for constructing a material database based on high-throughput experimental multimodal data according to claim 1, characterized in that: The sample coding system includes assigning a globally unique sample identifier to each catalyst sample and recording the sample's chemical composition, preparation method, and channel position coordinate information in the screening device; the multi-device time synchronization mechanism includes using the network time protocol to synchronize the infrared thermal imager, temperature sensor, mass flow controller, pressure sensor, and central control system at the millisecond level before the experiment; the adaptive acquisition frequency parameters of the infrared thermal imager include reducing the acquisition frequency during the stable reaction phase and increasing the acquisition frequency during the temperature mutation phase.

3. The method for constructing a material database based on high-throughput experimental multimodal data according to claim 1, characterized in that: The synchronously collected and recorded experimental metadata includes: unique experiment identifier, experiment date, experimenter information, experimental device identification, sample batch information, experiment batch information, temperature program setting value, target flow rate and reaction pressure of each gas component; the infrared thermal imaging data includes the original temperature matrix data recording the temperature changes of the catalyst channel; the sensor time series data includes the platform thermocouple temperature value, gas flow value and pressure value.

4. The method for constructing a material database based on high-throughput experimental multimodal data according to claim 1, characterized in that: The image preprocessing includes geometric correction of the infrared image sequence; the channel calibration includes using a pre-stored standard channel template image of the screening device to align each frame of the infrared image with the template through image registration to accurately segment the region of interest of each channel; The feature calculation includes extracting time series and spatial distribution features of the temperature data in the region of interest of each channel.

5. The method for constructing a material database based on high-throughput experimental multimodal data according to claim 4, characterized in that: The temporal temperature characteristics include ignition time, ignition temperature, maximum temperature rise, peak temperature and its arrival time, average temperature in the reaction stable stage and temperature fluctuation standard deviation; the spatial temperature distribution characteristics include temperature standard deviation and hot spot area ratio in the region of interest.

6. The method for constructing a material database based on high-throughput experimental multimodal data according to claim 5, characterized in that: The ignition time in the time series temperature characteristics is calculated by normalizing the dynamic ignition criterion To determine, when The ignition time is determined when the preset ignition criterion threshold value is continuously exceeded for at least a preset time period, wherein Calculated by the following formula: ， in, is the average channel temperature at time i, is the average channel temperature at the previous moment, is the time interval between time i and time i-1, is the normalized temperature rise rate parameter, is the reference baseline temperature, is the characteristic activation temperature parameter, is the index adjustment factor.

7. The method for constructing a material database based on high-throughput experimental multimodal data according to claim 1, characterized in that: The time alignment includes accurately aligning infrared thermal imaging data with sensor time series data on the time axis based on high-precision timestamps; the standardization processing includes converting data from different sources into a unified international system of units, eliminating sensor failure noise points, extracting statistics of time-varying parameters in key reaction stages, and forming standardized records.

8. The method for constructing a material database based on high-throughput experimental multimodal data according to claim 1, characterized in that: The three-layer logical architecture includes a basic metadata layer, a derived feature data layer, and a raw data index layer; the hybrid physical storage strategy includes storing structured metadata and feature data through a relational database, and storing large-capacity raw infrared thermal imaging video files and sensor log files through a distributed object storage system.

9. The method for constructing a material database based on high-throughput experimental multimodal data according to claim 8, characterized in that: The basic metadata layer includes a sample information table, a preparation method table, an experimental configuration table, and a channel mapping table; the derived feature data layer includes a catalyst performance table and a time series data summary table; the raw data index layer includes a raw data registry and a key frame table; and a standardized application interface is provided to support data operations and integration.

10. A material database construction system based on high-throughput experimental multimodal data, for implementing the method according to any one of claims 1 to 9, characterized in that: The system comprises: The data protocol management module is used to establish a standardized experimental data acquisition protocol. It includes a sample encoding submodule for assigning unique identifiers to samples and recording their chemical composition, preparation method, and channel location information; a time synchronization submodule for achieving millisecond-level clock synchronization of experimental equipment; and an acquisition frequency control submodule for setting the adaptive acquisition frequency parameters of the infrared thermal imager. The multimodal data acquisition module is used to synchronously collect and record experimental metadata, infrared thermal imaging data, and sensor time series data according to the standardized experimental data acquisition protocol, and perform data integrity verification to generate a timestamped and verified original multimodal experimental dataset; The infrared image processing module is used to process the infrared thermal imaging data in the original multimodal experimental dataset. It includes an image preprocessing unit for geometric correction of the infrared image sequence; a channel calibration unit for accurately aligning the infrared image with the standard channel template and segmenting the channel area; and a feature calculation unit for extracting structured temperature features including temporal temperature features and spatial temperature distribution features. The data integration and association module is used to integrate the experimental metadata and sensor time series data in the original multimodal experimental dataset, and combine them with structured temperature features to generate standardized sample-experiment-performance association data records through time alignment and normalization. The database management module is used to build and maintain the material database based on standardized sample-experiment-performance association data records and index information of the original multimodal experimental data set, using a three-tier logical architecture and a hybrid physical storage strategy. The three-tier logical architecture includes a basic metadata layer, a derived feature data layer, and a raw data index layer.

Images