7 bài học từ doanh nghiệp sản xuất linh kiện khi dùng AI để giảm lỗi QC và tránh giao hàng sai chuẩn

7 Lessons from a Component Manufacturer Using AI to Reduce QC Errors and Prevent Non-Conforming Shipments

1. Chẩn đoán rủi ro cụ thể: lỗi QC thường không nằm ở tay nghề, mà nằm ở dữ liệu rời rạc Một chủ xưởng sản xuất linh kiện từng...

1. Specific risk diagnosis: QC errors often come from scattered data, not poor skills A component factory owner once told the HimiTek team: “My QC people are not lazy. But...

batch_id,part_code,machine_id,shift,measured_value,min_spec,max_spec,defect_type,inspector,image_url,supplier_id
LOT-2409-A,PN-8821,M03,night,10.08,9.95,10.05,dimension_ng,Nguyen A,https://qc.local/img001.jpg,SUP-12
LOT-2409-A,PN-8821,M03,night,10.02,9.95,10.05,ok,Nguyen A,,SUP-12
LOT-2409-B,PN-7710,M01,day,5.01,4.98,5.03,scratch,Tran B,https://qc.local/img021.jpg,SUP-07

import pandas as pd

qc = pd.read_csv("qc_shift.csv")

# Sai lệch thông số
qc["out_of_spec"] = (qc["measured_value"] < qc["min_spec"]) | (qc["measured_value"] > qc["max_spec"])

# Đếm lỗi theo lô và máy
risk_summary = qc.groupby(["batch_id", "machine_id"]).agg(
    total_checks=("part_code", "count"),
    out_of_spec_count=("out_of_spec", "sum"),
    defect_count=("defect_type", lambda x: (x != "ok").sum())
).reset_index()

# Quy tắc cảnh báo thực dụng
risk_summary["risk_level"] = "low"
risk_summary.loc[risk_summary["out_of_spec_count"] >= 1, "risk_level"] = "verify"
risk_summary.loc[risk_summary["defect_count"] >= 3, "risk_level"] = "high"

# Danh sách cần trưởng ca xem trước khi xuất kho
hold_list = risk_summary[risk_summary["risk_level"].isin(["verify", "high"])]
print(hold_list)

batch_id,part_code,machine_id,shift,measured_value,min_spec,max_spec,defect_type,inspector,image_url,supplier_id
LOT-2409-A,PN-8821,M03,night,10.08,9.95,10.05,dimension_ng,Nguyen A,https://qc.local/img001.jpg,SUP-12
LOT-2409-A,PN-8821,M03,night,10.02,9.95,10.05,ok,Nguyen A,,SUP-12
LOT-2409-B,PN-7710,M01,day,5.01,4.98,5.03,scratch,Tran B,https://qc.local/img021.jpg,SUP-07

import pandas as pd

qc = pd.read_csv("qc_shift.csv")

# Out-of-spec measurements
qc["out_of_spec"] = (qc["measured_value"] < qc["min_spec"]) | (qc["measured_value"] > qc["max_spec"])

# Count issues by batch and machine
risk_summary = qc.groupby(["batch_id", "machine_id"]).agg(
    total_checks=("part_code", "count"),
    out_of_spec_count=("out_of_spec", "sum"),
    defect_count=("defect_type", lambda x: (x != "ok").sum())
).reset_index()

# Practical warning rules
risk_summary["risk_level"] = "low"
risk_summary.loc[risk_summary["out_of_spec_count"] >= 1, "risk_level"] = "verify"
risk_summary.loc[risk_summary["defect_count"] >= 3, "risk_level"] = "high"

# List for shift leader review before shipment
hold_list = risk_summary[risk_summary["risk_level"].isin(["verify", "high"])]
print(hold_list)