目的 旨在克服传统印刷质量评价中主观性偏差与指标权重设定争议,构建一套客观、可量化的多维度评价体系,为印刷工艺参数的精准调控与优化提供科学依据与决策支持。方法 基于36个实际印刷品样本数据（包含实地密度、相对反差、印刷光泽度、网点扩大、叠印率等指标）,采用五级标度法定义质量等级。采用有序Logistic回归（OLR,α=0.05,最大迭代次数=500）进行质量等级分类并量化参数影响（计算回归系数β、优势比OR及其95%置信区间）;利用随机森林（RF,树数=500,最大深度=10）评估特征重要性（基于Gini不纯度减少量）;构建贝叶斯层次模型（BHM,MCMC采样3 000次,预热1 000次,链数=4）以捕捉非线性交互效应及样本异质性。整合OLR、RF与BHM构建协同模型,并通过准确率、AUC及综合质量评分（CQS）进行模型验证。结果 OLR确定实地密度（β=0.82, OR=2.27, P<0.001）、相对反差（β=0.93, OR=2.53, P<0.001）和网点扩大（β=0.57, OR=1.77, P<0.001）为核心正向预测因子。随机森林特征重要性分析结果显示,实地密度（重要性权重0.31）对印刷质量影响最为显著,其次为相对反差（0.25）与网点扩大（0.22）,三者累计贡献度达78%,进一步验证了其在质量控制中的核心地位。BHM证实了实地密度（后验均值为0.80, 95% HDI [0.65, 0.95]）和相对反差（后验均值为0.91, 95% HDI [0.73, 1.09]）的主效应及其显著的交互作用（β=0.42）。协同模型的整体准确率达到84.7%,较单一OLR模型（78.3%）提升了6.4%,且对优秀等级样本表现出优异的区分能力（AUC=0.88）。综合质量排名与CQS呈显著正相关（r≈0.82）,同时更侧重于参数间的协同效应。结论 BHM-OLR-RF协同框架融合了多种模型的优势,显著提升了印刷质量评价的客观性、准确性与可解释性,精准量化了核心参数的影响及其重要性,有效解决了传统评价方法中存在的主观性与权重争议问题,为印刷工艺的多参数协同优化及力学性能改进提供了科学依据与决策支持。
Abstract
To address subjectivity and indicator weighting controversy in traditional print quality evaluation, the work aims to establish an objective and quantifiable multi-dimensional evaluation system, providing a scientific basis and decision-making support for the precise adjustment and optimization of printing process parameters. Based on the data from 36 print samples (including key indicators such as solid density, relative contrast, print gloss, dot gain, trapping rate, etc.), the quality was classified with a five-level scale. Ordinal Logistic Regression (OLR, α=0.05, max iterations=500) was adopted to classify quality grades and quantify parameter impacts (β, OR, 95% CI). The Random Forest (RF, trees=500, max depth=10) was employed to assess the feature importance (Gini decrease). A Bayesian Hierarchical Model (BHM, MCMC samples=3 000, warmup=1 000, chains=4) was constructed to capture non-linear interactions and heterogeneity. The OLR-RF-BHM model was established and validated via accuracy, AUC, and comprehensive quality ranking (CQS). OLR identified solid density (β=0.82, OR=2.27, P<0.001), relative contrast (β=0.93, OR=2.53, P<0.001), and dot gain (β=0.57, OR=1.77, P<0.001) as core positive predictors. RF importance analysis indicated that solid density (importance weight of 0.31) had the most significant impact on print quality, followed by relative contrast (0.25) and dot gain (0.22). The cumulative contribution of these three factors reached 78%. BHM confirmed main effects of solid density (post. mean 0.80, 95% HDI [0.65, 0.95]) and relative contrast (post. mean 0.91, 95% HDI [0.73, 1.09]) and their significant interaction (β=0.42). The synergistic model achieved 84.7% accuracy, a 6.4% improvement over OLR alone (78.3%), with superior discrimination for excellent grade (AUC=0.88). Comprehensive ranking correlated significantly with CQS (r≈0.82) but emphasized parameter synergy. The BHM-OLR-RF synergistic framework integrates model strengths, significantly enhancing objectivity, accuracy, and interpretability of print quality evaluation. It precisely quantifies core parameter impacts and importance, resolving subjectivity and weighting issues in traditional methods and providing a scientific basis and decision support for multi-parameter process optimization and mechanical improvement.
关键词
印刷质量评价 /
贝叶斯层次模型 /
有序 Logistic 回归 /
随机森林
Key words
print quality evaluation /
Bayesian Hierarchical Model (BHM) /
Ordinal Logistic Regression (OLR) /
random forest
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 张圣明, 孙倩倩, 崔晓萌, 等. 印刷线条质量的主观评价和分析[J]. 印刷技术, 2015(17): 43-45.
ZHANG S M, SUN Q Q, CUI X M, et al.Subjective Evaluation and Analysis of Printing Line Quality[J]. Printing Technology, 2015(17): 43-45.
[2] 何自芬, 张印辉. 纸张印刷适性综合评估体系的建立[J]. 印刷质量与标准化, 2004(7): 14-16.
HE Z F, ZHANG Y H.Foundation of the Paper Printability Comprehensive Assessment System[J]. Printing Quality & Standardization, 2004(7): 14-16.
[3] 张小雪, 孙帮勇, 谭家海, 等. 印刷品缺陷检测方法综述[J]. 印刷与数字媒体技术研究, 2025(2): 1-10.
ZHANG X X, SUN B Y, TAN J H, et al.A Review of Defect Detection Methods for Printed Products[J]. Printing and Digital Media Technology Study, 2025(2): 1-10.
[4] 周子镱, 董武, 张艳, 等. 屏幕内容图像质量客观评价方法综述[J]. 印刷与数字媒体技术研究, 2025(1): 14-27.
ZHOU Z Y, DONG W, ZHANG Y, et al.Review of Objective Quality Assessment Methods for Screen Content Images[J]. Printing and Digital Media Technology Study, 2025(1): 14-27.
[5] 孟唯娟, 张楷, 李治坤, 等. 不同纸张喷墨印刷质量评价方法的比较研究[J]. 印刷与数字媒体技术研究, 2025(1): 118-126.
MENG W J, ZHANG K, LI Z K, et al.Comparative Study on Evaluation Methods of Ink-Jet Printing Quality for Different Kinds of Paper[J]. Printing and Digital Media Technology Study, 2025(1): 118-126.
[6] 朱大鹏, 余珍, 曹兴潇. 基于贝叶斯推理的包装件动力学模型优化选择研究[J]. 包装工程, 2023, 44(5): 238-243.
ZHU D P, YU Z, CAO X X.Kinetic Package Model Optimal Selection Using Bayesian Inference[J]. Packaging Engineering, 2023, 44(5): 238-243.
[7] 陈子昂, 徐娟芳, 刘征. 考虑eHMI的行人过街行为研究与决策模型构建[J]. 包装工程, 1-10[2026-03-01]. https://link.cnki.net/urlid/50.1094.TB.20230615.1726.006.
CHEN Z A, XU J F, LIU Z. Pedestrian Crossing Behavior Research and Decision-Making Model Construction Considering eHMI[J]. Packaging Engineering, 1-10[2026-03-01]. https://link.cnki.net/urlid/50.1094.TB.20230615.1726.006.
[8] 刘战豫, 张宇飞. 基于机器学习的区域快递包装需求预测优化研究[J]. 包装工程, 2024, 45(1): 246-253.
LIU Z Y, ZHANG Y F.Optimization of Regional Express Packaging Demand Prediction Based on Machine Learning[J]. Packaging Engineering, 2024, 45(1): 246-253.
[9] 唐万有. 印刷品质量评价方法[J]. 印刷世界, 2004(2): 21-24.
TANG W Y.How to Appraise the Quality of Press Product[J]. Print World, 2004(2): 21-24.
[10] 王凯, 张彦. 基于模糊灰关联的纸张印刷适性评价模型[J]. 包装工程, 2016, 37(19): 195-200.
WANG K, ZHANG Y.Analysis of Paper Evaluation Model Based on Fuzzy Grey Relation[J]. Packaging Engineering, 2016, 37(19): 195-200.
[11] 王凯, 张彦. 基于熵权-云模型的印品质量评价模型[J]. 包装工程, 2018, 39(19): 244-250.
WANG K, ZHANG Y.Print Quality Evaluation Model Based on Entropy Weight-Cloud Model[J]. Packaging Engineering, 2018, 39(19): 244-250.
[12] 陈晓军, 王启明. 基于有序Logistic回归模型的通风系统评价[J]. 煤矿机械, 2024, 45(10): 192-195.
CHEN X J, WANG Q M.Evaluation of Ventilation System Based on Ordered Logistic Regression Model[J]. Coal Mine Machinery, 2024, 45(10): 192-195.
[13] 王世涛, 陈文思. 老年人体育参与影响因素的有序多分类Logistic回归分析[J]. 中国老年学杂志, 2024, 44(16): 4068-4072.
WANG S T, CHEN W S.An Ordered Multi-Classification Logistic Regression Analysis on the Influencing Factors of Sports Participation of the Elderly[J]. Chinese Journal of Gerontology, 2024, 44(16): 4068-4072.
[14] 付海涛, 张智勇, 王增辉, 等. 改进SHO算法优化随机森林模型[J]. 吉林大学学报(理学版), 2025, 63(3): 861-866.
FU H T, ZHANG Z Y, WANG Z H, et al.Improve SHO Algorithm to Optimize Random Forest Model[J]. Journal of Jilin University (Science Edition), 2025, 63(3): 861-866.
[15] 孙静, 邹向坤, 朱晓勃, 等. 随机森林算法在网络号贩子预测领域的研究[J]. 计算机仿真, 2025, 42(7): 549-553.
SUN J, ZOU X K, ZHU X B, et al.Research on Random Forest Algorithm in Field of Internet Scalper Prediction[J]. Computer Simulation, 2025, 42(7): 549-553.
基金
温州市科技项目（2025R0266）; 浙江省教代会工会研究课题（ZJJYGHYB2025050）
PDF(636 KB)
渝公网安备 50010702501716号 渝ICP备15012534号-2
