2nd K-AI Manufacturing Competition (0)

▤ 목차

• Overview
• Analysis Background
  • Overview of the Process and Equipment
  • Pain Points in the Current Process
  • Problems to Address
• Analysis Objectives
  • Key Objective
  • Purpose of Manufacturing Data Analysis

Overview

KAMP, an AI manufacturing platform managed by the Ministry of SMEs and Startups in the Republic of Korea, held a competition. The goal was to define and solve a problem based on the anonymized random dataset, and we received a melting tank dataset from the food manufacturing industry.

728x90

Analysis Background

반응형

Overview of the Process and Equipment

The dataset analyzed in this study originates from the melting tank process, a critical preprocessing stage in the production of powdered cream within the food manufacturing industry. This process entails dissolving powdered raw materials into liquid components within spray-drying production facilities. As the initial step in raw material preparation, this mixture's uniformity directly impacts the final product's quality.

 

Key operational parameters, such as temperature and stirring speed, significantly influence the success of the melting process. Insufficient temperatures or slow stirring speeds can lead to incomplete dissolution, compromising downstream processes. Conversely, excessively high temperatures or overly rapid stirring may trigger undesirable chemical reactions, impacting product quality and potentially causing wear on equipment. Furthermore, a range of external factors beyond these variables also affect the performance of the melting process.

 

Key Takeaway:

Optimizing the melting process is essential to achieving consistent product quality, ensuring the downstream production stages operate smoothly, and minimizing risks to equipment longevity.

Pain Points in the Current Process

The melting process, as the first stage in raw material preparation, profoundly impacts subsequent production steps and the quality of the final product. In powdered cream manufacturing, the process relies heavily on spray-drying techniques, with the melting stage setting the foundation for all subsequent drying and production phases. If the melting process does not achieve proper dissolution, it becomes increasingly challenging to guarantee the quality of the powdered end product.

 

Several specific challenges have been identified:

• Operational Complexity: When multiple raw materials are added or large-scale production is undertaken, raw materials are often introduced sequentially. Each addition alters the internal conditions of the tank, such as temperature, viscosity, and stirring speed. These fluctuations require continuous manual adjustments to maintain process stability without clear operational guidelines.
• Reliance on Expertise: Adjustments to operating conditions are typically based on the experience and intuition of skilled operators. However, any gaps in personnel availability often result in difficulties in maintaining consistent product quality.
• External Variables: Factors such as seasonal or weather-related changes in humidity and temperature can impact raw materials and the tank environment. These variables are not always controllable, adding further complexity to process management.

Key Takeaway:

The absence of standardized guidelines and heavy reliance on skilled operators expose the melting process to significant variability, making it difficult to achieve consistent quality in large-scale or multi-ingredient operations.

Problems to Address

The primary issue with the current melting process is the lack of an effective, standardized method for measuring the dissolution state in real-time. Without this, ensuring quality during production becomes highly dependent on manual intervention and post-process evaluations.

 

Current challenges include:

• Delayed Quality Assessments: Quality inspections are conducted intermittently during production or only after downstream processes are complete, making it difficult to identify and address issues in real-time. This delay often leads to inefficiencies and suboptimal product quality.
• Subjective Quality Checks: The process relies heavily on operators' expertise to evaluate the state of dissolution, which introduces subjectivity and inconsistency. In many cases, there is no real-time monitoring of critical quality indicators, further hindering proactive adjustments during production.
• Inadequate Systems for Quality Assurance: The lack of an integrated system to monitor dynamic operational parameters, such as temperature, stirring speed, and material levels, in real-time makes it challenging to guarantee consistent product quality.

Proposed Solution:

To address these issues, we propose the development of a predictive system that models the relationship between operational parameters (e.g., temperature, stirring speed, material volume) and product quality in real-time. By leveraging this model, it becomes possible to anticipate and ensure product quality throughout the production process without relying solely on operator expertise or post-process inspections.

 

Key Takeaway:

Building a real-time predictive monitoring system can mitigate the reliance on subjective judgments, reduce quality assurance delays, and enable proactive control of the melting process to consistently meet production standards.

Analysis Objectives

Key Objective

The primary objective of this study is to develop an analytical model capable of predicting the quality of final products during the melting process in real time. By addressing the challenges identified in Section 1, the model leverages data collected via programmable logic controllers (PLCs) to optimize operational parameters and ensure consistent product quality.

Purpose of Manufacturing Data Analysis

To tackle the outlined challenges, the melting process is reframed as an anomaly detection problem within the manufacturing sector. Specifically, the model uses the dataset's defect label ("TAG") as the target variable to train a supervised learning model.

 

Given the complexity of the production process, multiple factors can influence product quality. The dataset includes key variables critical to this process, such as temperature, stirring speed, and material levels. These are incorporated into the model to ensure comprehensive analysis.

 

Key Considerations:

• Imbalanced Data Handling: Since defective products (labeled "NG") comprise a small proportion of the dataset, the model applies a synthetic oversampling technique (SMOTE) based on the K-nearest neighbors method. This ensures balanced training and robust predictions.
• Goal: To develop a predictive model that utilizes all relevant variables to accurately classify and predict product quality during the melting process.

Key Takeaway:

The project aims to deliver a data-driven solution that reduces reliance on manual interventions, enhances process efficiency, and ensures consistent product quality by proactively identifying and addressing quality anomalies.