Molecular comparison system and method thereof

Abstract

According to the molecular comparison system and the molecular comparison method provided by the invention, after a substance analysis result is obtained, according to the molecular list, the matching degree numerical value and related problems and corresponding weights in the knowledge management system, the relevance database is used, and the AI LLM is used for judgment and calculation, so that the molecular type of a compound causing damage is further determined.

Description

Technical Field

[0001] The present invention generally relates to systems and methods for molecular alignment and, more particularly, to systems and methods for analyzing potentially damaging molecules using a knowledge management system and a large language model. Background Art

[0002] Nowadays, wearable devices such as smart bracelets cannot be locked and sealed with traditional screw reinforcement technology due to their small size. Instead, they are glued together to achieve the tightness and waterproofness of the entire device. However, after the glue is cured, it may be chemically degraded by chemicals in the environment, causing the glue to fail, thereby destroying the tightness and waterproofness of the entire device. In this case, testing is required to determine the specific type of chemical molecule that causes the glue to fail.

[0003] Among the many methods of substance analysis known to be used, FTIR (Fourier transform infrared) spectroscopy is used as an example. Although the type of molecular substances that may cause chemical damage can be inferred through absorption spectra, FTIR can only provide general guidance on the functional groups of molecules, which is insufficient for inferring the type of substances that cause damage. Therefore, there is a need for a system and method for determining molecules that cause damage. Summary of the invention

[0004] In the first embodiment, the direction of the present disclosure is a molecular comparison method. The molecular comparison method includes obtaining multiple questions related to material destruction and multiple weights corresponding to the multiple questions from a knowledge management system based on multiple molecular species in a molecular list. The molecular list contains multiple matching degree values ​​corresponding to multiple molecular species. The molecular comparison method also includes forming multiple AI question words through multiple molecular species and multiple questions in the molecular list. The molecular comparison method also includes inputting multiple AI question words into at least one LLM and generating corresponding multiple AI answers. The molecular comparison method also includes calculating according to multiple AI answers, multiple weights corresponding to multiple questions, and multiple matching degree values ​​corresponding to multiple molecular species to obtain multiple weight scores corresponding to multiple molecular species. According to the multiple weight scores, it is judged which of the multiple molecular species is a destructive molecule.

[0005] In a second embodiment, the direction of the present disclosure is a molecular comparison system. The molecular comparison system includes a material analysis module for generating a molecular list according to the test object, and the molecular list includes multiple molecular species and multiple matching degree values ​​corresponding to the multiple molecular species. The molecular comparison system also includes a knowledge management system, which includes multiple questions related to material destruction and multiple weights corresponding to the multiple questions. The molecular comparison system also includes a correlation database for obtaining multiple questions and multiple weights corresponding to the multiple questions from the knowledge management system according to the multiple molecular species in the molecular list, and forming multiple AI question words according to the multiple molecular species and multiple questions in the molecular list. The molecular comparison system also includes at least one LLM for receiving multiple AI question words and generating corresponding multiple AI answers. The molecular comparison system also includes a weighted calculation module for calculating according to multiple AI answers, multiple weights corresponding to multiple questions, and multiple matching degree values ​​corresponding to multiple molecular species to obtain multiple weight scores corresponding to multiple molecular species. The molecular comparison system also includes a judgment module for judging which of the multiple molecular species is a destructive molecule according to multiple weight scores.

[0006] In order to better understand the above and other aspects of the present invention, embodiments are given below and described in detail with reference to the accompanying drawings: BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 Schematic diagram showing an example molecular alignment system according to certain aspects of the present disclosure;

[0008] FIG. 2A to FIG. 2B is a schematic diagram of an example molecular inventory according to certain aspects of the present disclosure;

[0009] Figure 3 is a flow chart of an example procedure for molecular alignment according to certain aspects of the present disclosure.

[0010] Reference numerals

[0011] 100: System QS: Question

[0012] 110: Material Analysis Module Q AI :AI question words

[0013] 120: Relational Database RA AI : AI Answers

[0014] 130: Knowledge Management System RI: Related Information

[0015] 140: RAG architecture Rs: comparison results

[0016] 141: LLM SP: Object under test

[0017] 142: Retriever WS: Weight

[0018] 143: Private Domain Database WR: Weight Score

[0019] 150: Weighted calculation module 300: Program

[0020] 160: Judgment module S310, S320, S330, S340, S350: Steps

[0021] HQI: matching index

[0022] ML: Molecular List DETAILED DESCRIPTION

[0023] Figure 1 A schematic diagram of an example molecular comparison system 100 according to specific aspects of the present disclosure is shown. The system 100 may include a material analysis module 110, which may be used to generate a molecular list ML based on a test object SP, such as a test object SP sampled around a damaged substance (e.g., a substance that causes damage), such as a molecular list generated by similarity comparison through an FTIR spectrum database. This example takes the comparison to obtain the molecular types that damage reactive polyurethane hot melt adhesive (Polyurethane Reactive, PUR) glue as an example, but is not intended to be limiting. In this example, after obtaining the test object around the damaged PUR glue, the molecular list ML can be obtained through the material analysis module 110, such as Figure 2A The molecular list ML shown. The molecular list ML includes multiple names (Name) representing different molecular species, and multiple matching degree values ​​(HQI) corresponding to the multiple molecular species. In the molecular list ML of this example, possible molecular species (destructive substances-) include L-Lysine (HQI is 85.37%), L-Glutamic acid, magnesium salt, trihydrate (HQI is 84.65%), Monaquest IA (HQI is 84.55%), N-Hippuryl-His-Leuhydrate (HQI is 84.31%), Zirconium carbonate, Hydrodrated (HQI is 84.29%), etc.

[0024] The system 100 may include a knowledge management system 130 including, for example, a plurality of questions QS related to material damage and a plurality of weights WS corresponding to the plurality of questions, such as shown in Table 1 below.

[0025]

[0026]

[0027] (Table 1)

[0028] In some embodiments, the weight in the knowledge management system 130 may be represented by a plurality of values, each corresponding to a different answer to a question. In this example, the binary value of the weight is represented by 1 or 2, for example, 1 indicates no relevance to the corresponding question, and 2 indicates relevance to the corresponding question. In other embodiments, the weight may have three or more different values ​​to indicate the degree of relevance to the corresponding question. For example, the question "Does it exist in the daily life environment" may have a weight between 0 and 4. A weight of 0 indicates no existence, a weight of 1 indicates 25% possible existence, a weight of 2 indicates 50% possible existence, a weight of 3 indicates 75% possible existence, and a weight of 4 indicates 100% existence.

[0029] In some embodiments, multiple questions QS related to material damage of the knowledge management system 130 can be associated with glue failure analysis, and the multiple questions QS belong to multiple attributes, including environmental sources, product sources, and pH values, as shown in Table 1.

[0030] The system 100 may include a relational database 120, which may be used to obtain a plurality of questions QS and corresponding weights WS from a knowledge management system 130 according to a plurality of molecular species in the molecular list ML, and to form a plurality of AI question words Q according to the plurality of molecular species in the molecular list ML and the plurality of questions QS. AI In some embodiments, the AI ​​question words including the molecular species (for example, the top 5 molecular species with the highest HQI in the molecular list ML) and multiple questions can be represented by a matrix table. The system 100 may include at least one LLM 141. In this example, the AI ​​question word Q is input. AI The RAG (Retrieval-Augmented Generation) architecture 140 including the LLM 141 is taken as an example, but is not limited thereto. The RAG architecture 140 may include a retriever 142 for receiving an AI question word Q AI , and includes a private domain database 143, which is used by the retriever 142 according to the AI ​​question word Q AI Search and provide information about AI question words Q AI The private domain database 143 may include chemical resistance manuals, such as Kuraray's Chemical Resistance Guide for Kuraray Resin, American Chemical Society, and Elsevier and other chemical journals, but is not limited thereto, to provide AI question words QAI The RAG framework 140 can help ensure the accuracy of subsequent LLM answers, thereby reducing the generation of AI hallucinations (answers are more general and often irrelevant to the question). After obtaining the relevant data RI, the AI ​​question word Q AI Input the relevant information RI into LLM 141 to get the AI ​​answer RA AI In some embodiments, the AI ​​question word Q AI Input multiple LLMs with relevant information RI, or just input AI question word Q AI Input multiple LLMs to get AI answers provided by multiple different LLMs to increase the accuracy of the AI ​​answers.

[0031] The system 100 may include a weighted calculation module 150 for calculating the weighted value of the AI ​​answer RA according to the obtained AI answer RA. AI , multiple weights WS corresponding to multiple questions QS, and multiple matching degree values ​​HQI corresponding to multiple molecular species (included in the molecular list ML) are calculated to obtain multiple weight scores W corresponding to multiple molecular species R For example, the weighted calculation module 150 may generate different weight scores according to AI answers with different LLMs (LLM 1 and LLM 2), as shown in Table 2 and Table 3 below.

[0032]

[0033] (Table 2)

[0034]

[0035] (Table 3)

[0036] The AI ​​answer generated by the LMM may have any suitable format. In some embodiments, the indication may be a value corresponding to the weight of the question, such as the value of the weight corresponding to the question discussed above. In the examples of Tables 2 and 3, the weighted calculation module 150 calculates the weight score based on the HQI and the weight corresponding to the AI ​​answer to each question. For example, in Table 2, the weight score of L-Lysine is the sum of the HQI multiplied by the weight corresponding to each AI answer (1365.92=85.37x(2+2+2+2)).

[0037] The system 100 may include a judgment module 160, which may judge which of the multiple molecular species is a destructive molecule according to the weight score WR, and output a comparison result Rs. For example, the judgment module 160 may judge which is a destructive molecule according to the weight score of Table 2, or the weight score of Table 3, or the weight scores of both Table 2 and Table 3. For example, in Table 3, the weight score of L-Lysine is the highest (1365.92), so the destructive molecule can be judged to be L-Lysine, that is, in this example, the destructive molecule that causes the failure of the PUR glue is judged to be L-Lysine. The comparison result Rs can be output to a display device (not shown) to display it to the user, or stored as data in a medium, such as a storage device or non-volatile memory that can store data.

[0038] Next, the analysis of substances that damage the conductive contacts (contact pads) of PCB (printed circuit board) by the molecular comparison system 100 provided in the present disclosure will be taken as an example. In this example, after obtaining the test object SP around the damaged conductive contacts (causing the abnormal conductive contacts), the molecular list ML can be obtained through the material analysis module 110. For example, the molecular list generated by similarity comparison can be performed through energy dispersive X-ray spectroscopy (EDS) and gas chromatography mass spectrometry (GCMS) in conjunction with the FTIR spectral database. For example, after using a scanning electron microscope (SEM) to find the possible location of the destructive substance, the test object SP is analyzed with EDS in conjunction with the FTIR spectral database to obtain the molecular list ML, for example. Figure 2B The molecular list ML shown. EDS can also obtain the elemental composition percentages corresponding to the molecular list, similar to the aforementioned HQI. For example, in some embodiments, the elemental composition percentages of EDS can be used to replace HQI alone, or used together with HQI to calculate the weight score. In the molecular list ML of this example, possible molecular types (destructive substances-) include 2-ETHYLHEXANAL (HQI is 82.24%), Hexanal (HQI is 81.65%), Cyclopentanecarboxaldehyde, 2-hexyl-, trans- (HQI is 79.95), etc. Next, multiple questions QS related to material destruction in the knowledge management system 130 and multiple weights WS corresponding to the multiple questions can be used, as shown in Table 1 below.

[0039]

[0040] (Table 4)

[0041] Next, as described above, the association database 120 can be used to Figure 2BThe multiple molecular species in the molecular list ML obtain multiple questions QS and corresponding multiple weights WS (for example, from Table 4) from the knowledge management system 130, and form multiple AI question words Q according to the multiple molecular species in the molecular list ML and the multiple questions QS AI .

[0042] In this example, we also input the AI ​​question word Q AI Taking the RAG (Retrieval-Augmented Generation) architecture 140 including the LLM 141 as an example, the relevant information RI of the AI ​​question word QAI is provided, and after obtaining the relevant data RI, the AI ​​question word QAI can be AI Input the relevant information RI into LLM 141 to get the AI ​​answer RA AI .

[0043] Next, the AI ​​question word Q AI The LLM weight calculation module 150 is input, and the weight calculation module 150 is used to calculate the weight of the AI ​​answer RA according to the obtained AI answer RA. AI , multiple weights WS corresponding to multiple questions QS, and multiple matching degree values ​​HQI corresponding to multiple molecular species (included in the molecular list ML) are calculated to obtain multiple weight scores W corresponding to multiple molecular species R , as shown in Table 5 below (only three molecules with higher weight scores are listed).

[0044]

[0045] (Table 5)

[0046] The determination module 160 may determine, according to the weight scores in Table 5, that Abietic acid (312.2) having the highest weight score is the destructive molecule causing the poor conductive contact.

[0047] In another example, the molecular comparison system 100 provided by the present disclosure can analyze the destructive substances that cause the circuit pins to break. Similarly, after obtaining the object to be tested SP around the broken circuit pins, the molecular list ML can be obtained through the material analysis module 110. For example, after using a scanning electron microscope (SEM) to find the possible location of the destructive substance, the object to be tested SP is analyzed by EDS in conjunction with the FTIR spectrum database to obtain the molecular list ML and the elemental composition percentage. The multiple questions QS related to material damage and the multiple weights WS corresponding to the multiple questions in the knowledge management system 130 may be related to the elements contained in the pins themselves (Fe, Co, Ni, etc.). For example, if it is an element contained in the pins themselves, the weight WS is lower, otherwise the weight WS is set higher. Then, through a procedure similar to the above example (omitted here), it can be concluded that chlorine (Cl) with the highest weight score is the destructive molecule that causes the circuit pins to break.

[0048] Figure 3 FIG. 3 is a flow chart of an exemplary procedure 300 for molecular alignment according to certain aspects of the present disclosure. In step 310, according to a list of molecules of a substance to be analyzed obtained from a substance analysis module (e.g., Figure 1 The multiple molecular species in the molecular list ML obtained by the substance analysis module 110 of the test substance SP are obtained from the knowledge management system (eg Figure 1 The knowledge management system 130 of the invention obtains multiple questions related to material damage (such as PUR glue or conductive contacts of PCB) and multiple weights corresponding to the multiple questions. The molecular list includes multiple matching degree values ​​(HQI) corresponding to multiple molecular species. In step 320, for example, Figure 1 The associative database 120 forms multiple AI question words (e.g. Figure 1 AI question word Q AI In step 330, multiple AI question words are input to at least one LLM (e.g. Figure 1 LLM 141), and generates corresponding multiple AI answers (such as Table 2, Table 3, Table 5 or Figure 1 AI answers RA AI In some embodiments, as discussed above, the AI ​​answer can be obtained by inputting the AI ​​question words into a RAG framework (e.g. Figure 1 In step 340, for example, Figure 1 The weighted calculation module 150 performs calculations based on the multiple AI answers, the multiple weights corresponding to the multiple questions, and the multiple matching degree values ​​corresponding to the multiple molecular species to obtain multiple weight scores corresponding to the multiple molecular species (e.g., Table 2, Table 3, Table 5 or Table 6). Figure 1 The weight score WR ). In step S350, for example, Figure 1 The judgment module 160 judges which of the multiple molecular species is a destructive molecule according to multiple weight scores, for example, by comparing the weight scores of each molecular species, the molecular species with the largest weight score is judged as a destructive molecule. The program 300 can be implemented in whole or in part on any suitable computing device, such as the system 100 or other computing devices. For example, in some cases, steps S320, S340, and S350 can be executed using the system 100 to determine the destructive molecule. Step S310 (obtaining multiple questions related to material destruction and multiple weights corresponding to the multiple questions) and step S330 (inputting multiple AI question words into the LLM) can be executed on the system 100 or on another computing device, such as other computing devices connected through a network.

[0049] According to the technology provided by the various embodiments of the present disclosure, it is possible to effectively determine which of the multiple molecular species is a destructive molecule, which can be used to improve products (such as wearable devices), replace the possibility of conventional molecules, and increase the accuracy of verification. For example, by determining the specific destructive molecule that causes the failure of PUR glue, it can help to replace the ingredients of the glue, or develop new glue, or determine the specific destructive molecule that causes poor conductive contacts on PCBs, thereby improving the durability and reliability of the product.

Claims

1. A molecular comparison method, characterized in that: include: Obtaining multiple questions related to material destruction and multiple weights corresponding to the questions from a knowledge management system according to multiple molecular species in a molecular list, wherein the molecular list includes multiple matching degree values ​​corresponding to the molecular species; Form a plurality of AI question words using the molecular species in the molecular list and the questions; Input the AI ​​question words into at least one LLM and generate corresponding multiple AI answers; Calculating according to the AI ​​answers, the weights corresponding to the questions, and the matching degree values ​​corresponding to the molecular species to obtain a plurality of weight scores corresponding to the molecular species; and According to the weight scores, it is determined which of the molecular types are destructive molecules.

2. The molecular comparison method according to claim 1, characterized in that: The weights corresponding to the questions have a binary value for indicating whether the molecular species are relevant to the questions.

3. The molecular comparison method according to claim 1, characterized in that: The weights corresponding to the questions have three or more different values ​​for indicating the degree to which the molecular species are relevant to the questions.

4. The molecular comparison method according to claim 1, characterized in that: Inputting the AI ​​questions into the at least one LLM and generating the corresponding AI answers include: Input the AI ​​question words into a search engine; The searcher searches a private domain database for relevant information related to the AI ​​question words; and The relevant information and the AI ​​question words are input into the at least one LLM, and the corresponding AI answers are generated.

5. The molecular comparison method according to claim 1, characterized in that: The issues related to physical damage are associated with glue failure analysis and belong to multiple attributes, including environmental source, product source and pH value.

6. A molecular comparison system, characterized in that: include: A material analysis module, used for generating a molecule list according to a test object, wherein the molecule list includes a plurality of molecule types and a plurality of matching degree values ​​corresponding to the molecule types; a knowledge management system including a plurality of questions related to material damage and a plurality of weights corresponding to the questions; a correlation database, for obtaining the questions and the weights corresponding to the questions from the knowledge management system according to the molecular species in the molecular list, and forming a plurality of AI question words according to the molecular species in the molecular list and the questions; At least one LLM, configured to receive the AI ​​question words and generate corresponding AI answers; a weighted calculation module, for calculating according to the AI ​​answers, the weights corresponding to the questions, and the matching degree values ​​corresponding to the molecular species, so as to obtain a plurality of weight scores corresponding to the molecular species; and A judgment module is used to judge which of the molecular types is a destructive molecule according to the weight scores.

7. The molecular comparison system according to claim 6, characterized in that: The weights corresponding to the questions of the knowledge management system have binary values ​​for indicating whether the molecular species are related to the questions.

8. The molecular comparison system according to claim 6, characterized in that: The weights corresponding to the questions of the knowledge management system have three or more different values ​​for indicating the degree to which the molecular species are related to the questions.

9. The molecular comparison system according to claim 6, characterized in that: Also includes: A searcher, for receiving the AI ​​question words; and a private domain database, for being searched by the search engine according to the AI ​​question words, and providing relevant information about the AI ​​question words, The relevant information and the AI ​​question words are input into the at least one LLM, and the corresponding AI answers are generated.

10. The molecular comparison system according to claim 6, characterized in that: The questions related to material damage of the knowledge management system are associated with glue failure analysis, and the questions belong to multiple attributes, including environmental sources, product sources, and pH values.

Images