Engineering Application Practice of Artificial Intelligence in Manufacturing Industry--The Ninth Lecture of Industrial Software Forum

• artificial intelligence
  • New-Generation General Purpose Technology
    • Data-driven artificial intelligence is likely to become a general-purpose technology (General-Purpose Technology) as defined by economists. The impact is usually huge and far-reaching, just like the historical significance of the internal combustion engine, electricity, and the Internet.
  • Digital talent
  • Artificial intelligence technology landscape
  • Industrial digital twin built based on data science methods
  • Smart Manufacturing and Digital Twins
  • Scenarios that digital twins built with big data can serve
    • The full life cycle of manufacturing
      • Product development
      • manufacturing
      • supply chain management
      • sales channel management
      • User management
      • customer service
• The Practice of Artificial Intelligence in Manufacturing
  • Features of Industrial Big Data Application
    • Based on the product and process mechanism, it is necessary to form an agile, high-accuracy and generalizable prediction model by mining and integrating massive and multi-source data, so that it is possible to drive business innovation and mathematical transformation and upgrading of enterprises.
    • Systematic: It is necessary to integrate various related technologies to form an effective closed-loop from analysis to control. Agility: Quickly complete the creation, modeling, verification and deployment of business topics. Continuity: The data analysis model needs to have the ability to be generalized
    • Process: pay attention to problem-solving-oriented convergence process and process management to ensure the certainty of results
    • Standardization: Based on the standards of existing industrial systems, establish data standards, analysis standards, standards for cognition of analysis results, and standards for feedback operations.
  • Data Science Application Implementation Process
    • Method: Perform deductive reasoning and correlation induction on the production results and process parameters, and optimize the production process parameters through big data modeling to improve the efficiency and quality of the production line.
  • Algorithm model engineering method
    • application platform
    • Data center
    • Algorithm platform
  • The essence of algorithm encapsulation: cross-language, cross-platform interprocess communication
  • Data Science Tool 1: Machine Learning
    • Application Example: Production Control Optimization Digital Twin
    • Machine learning application example: digital twin of equipment operating conditions
    • Application example: digital twin of excavator working conditions
    • Excavator Data Algorithm Model Application Example
    • Machine Learning Application Example: Equipment Productivity Digital Twin
    • Application example: digital twin of working efficiency of rotary drilling rig
    • Machine Learning Application Example: Sales Forecasting
      • According to the sales volume of small excavators, medium excavators, large excavators and micro excavators over the years and the operating rate and operating hours to predict the sales volume of excavators in the next year, subdivide them into key prefecture-level cities and predict them separately, and calculate the specific sales volume at the same time Influencing factors, multi-dimensional and comprehensive analysis of market dynamics, and accurate prediction of market trends.
    • Data Columbus: The Data Science "Nuclear Weapon"
    • Machine Learning Application Example: Pre-loan Risk Assessment
      • Model selection:
      • CatBoost algorithm
      • Accuracy (F1): 85%
    • Machine Learning Application Example: Risk Monitoring in Loans
      • Model selection:
      • LightGBM
      • Algorithm accuracy (F1): 92%
    • Machine Learning Application Example: WIP Quality Digital Twin
    • Machine Learning Application Example: Lifespan Prediction Based on Image Features
  • Data Science Tool 2: Simulation Simulation
    • Simulation application example: digital twin for production takt optimization
    • Simulation application example: production cycle simulation
  • Data Science Tool 3: Operational Research Optimization
    • Operations Research
      • Operations research is the science of decision-making. Operations research can be used when decisions are made.
      • Decision-making is an irreversible resource allocation process. Therefore, when resource allocation occurs, operations research can be used
      • Find the optimal decision that maximizes/minimizes an objective subject to constraints
    • Operational optimization application example: production scheduling optimization
    • Operational optimization application example: intelligent logistics delivery
  • Data Science Tool 4: Knowledge Graph
    • Knowledge Map and NLP Application Example: Fault Diagnosis Expert System
      • Through NLP, deep learning and other technologies, information extraction of entities and relationships such as parts, fault phenomena, and maintenance methods is realized, and a knowledge map of excavator fault maintenance is constructed, and applications are carried out in semantic retrieval, problem classification, and prediction to realize fault modes. Mining and identification, solidifying expert knowledge and maintenance methods automatically identified by the system, and building an after-sales service expert system.
    • Knowledge Graph and NLP Application Examples: Big Data Analysis of Customer Needs
• enlightenment
  • Summary of experience in algorithm model packaging and application technology
  • Thoughts on Industrial Software R&D in the AI ​​Era
    • 01 Software relies on the market to grow
    • 02X factor
    • 03Value VS Ease of Use
  • The deep pit of artificial intelligence in industry
  • Conclusion: It is not the strongest, nor the smartest, but the species that can best adapt to changes, in order to survive. —Darwin