Onshore oil and gas pipelines face significant corrosion challenges due to environmental factors like soil conditions, moisture, and corrosive fluids. Common corrosion types include CO₂ corrosion, causing uniform thinning in sweet environments; H₂S corrosion, leading to sulfide stress cracking in sour service; microbiologically influenced corrosion (MIC), driven by bacterial activity in low flow areas; pitting corrosion, resulting in localized material loss; and stress corrosion cracking (SCC), worsened by tensile stresses and corrosive media [5]-[8]. These mechanisms threaten pipeline integrity, risking leaks or failures [9]. Material choices typically include carbon steels for cost effectiveness in mild conditions, stainless steels for improved resistance to pitting and general corrosion, and duplex steels for enhanced strength and corrosion resistance in aggressive environments with high chloride or H₂S content [5][6]. Brief description on basic of corrosion are presented in Table 1.

Traditional material selection approaches typically involve a three-step methodology: screening materials using constraints, ranking them using performance indices, and finally seeking supporting information for top candidates. These methods rely on structured data and include strategies such as quantitative analysis, questionnaire-based expertise capture, and case-based analogy [11]. Traditional

Table 1.Brief on basic types of corrosion [10].

material selection for corrosion resistant steels in pipelines relies on performance testing, such as immersion or electrochemical tests to measure corrosion rates, cost benefit analysis to balance initial and lifecycle costs, and empirical models based on standards like NACE MR0175 for sour service [12]. These methods evaluate alloy composition, mechanical properties, and environmental compatibility to select materials like carbon or stainless steels [13]. However, limitations include time intensive testing, data gaps from incomplete field histories, and subjectivity in interpreting results or tradeoffs, often leading to conservative designs or overlooked risks [14].

Performance Testing, Cost Benefit Analysis, Empirical Models

[16] presents an empirical evaluation of test case prioritization techniques within Junit based Java environments. The study demonstrates that prioritization significantly enhances fault detection rates, particularly for techniques incorporating feedback. Performance was measured via the APFD metric, with method level and block level techniques showing notable improvements over random or untreated test orders. A cost benefit analysis reveals that prioritization is most effective in incremental or continuous testing contexts, where frequent test executions amplify time savings. The authors introduce cost models differentiating between batch and incremental testing processes, highlighting that savings. Figure 1 shows the organization of structured data of the sort generally found in handbooks and data sheets. While the different stages for materials and process selection in design and the conceptual tools to carry them out are presented in Figure 2.

Figure 1.The organization of structured data of the sort generally found in handbooks and data sheets. The material or process is indexed by name; the information listed under the name gives numeric values for material properties, a ranking of performance under standard conditions and yes or no categorization of material–process compatibility [11].

Figure 2. The different stages for materials and process selection in design and the conceptual tools to carry them out [15].

depend on organizational cost structures and test suite characteristics. The empirical models used grounded findings in real world Java programs and seeded faults, offering practical insights into tradeoffs between prioritization overhead and early fault detection benefits. The study underscores the context dependent value of prioritization and provides a framework for assessing its applicability in industrial settings. a corrosion management case study comparing the lifecycle costs of carbon steel and corrosion-resistant alloy (CRA) pipelines in aggressive environments is also utilized. Although CRAs have higher initial material and installation costs, their superior resistance to corrosion significantly reduces maintenance, inspection, downtime, and failure-related expenses over the service life of the asset, resulting in a lower total lifecycle cost.

AI assisted material selection represents a transformative approach to identifying corrosion resistant steels for onshore oil and gas pipelines, leveraging advanced computational techniques to overcome the limitations of traditional methods. By integrating machine learning (ML) and deep learning (DL) algorithms, this approach enables rapid, data driven predictions of material performance under complex corrosion conditions, optimizing the selection process for alloys like carbon steels, stainless steels, and duplex steels. These methods excel at analyzing large, multidimensional datasets encompassing alloy composition, environmental conditions, and corrosion behavior, offering engineers predictive tools that enhance efficiency and accuracy [17][18].

2.3.1. Machine Learning and Deep Learning Methods

Several ML and DL techniques are employed in material selection for corrosion resistance, each with distinct strengths:

1)Artificial Neural Networks (ANNs): ANNs are widely used for their ability to model nonlinear relationships in complex datasets, such as predicting corrosion rates based on alloy composition and environmental factors (e.g., pH, temperature, CO₂/H₂S partial pressures). They consist of interconnected nodes that learn patterns through training on historical corrosion data, making them suitable for forecasting material performance in pipelines [17][19].

2)Support Vector Machines (SVMs): SVMs are effective for classifying materials as suitable or unsuitable for corrosive environments, such as sour service with high H₂S content. By mapping data into higher dimensional spaces, SVMs identify optimal boundaries for material performance, offering high accuracy in scenarios with limited training data [20]. They are particularly useful for binary decisions, such as selecting between stainless and duplex steels for pitting resistance [21].

3)Decision Trees and Ensemble Methods:Decision trees provide interpretable models by breaking down material selection into hierarchical decisions based on features like alloy microstructure or chloride exposure. Ensemble methods, such as random forests or gradient boosting, combine multiple trees to improve prediction robustness, making them ideal for handling heterogeneous pipeline corrosion data [22][23].

4)Deep Learning Models: DL models, such as convolutional neural networks (CNNs) or physics informed neural networks (PINNs), are increasingly applied to complex corrosion scenarios. CNNs can analyze microstructural images to correlate grain size or phase distribution with corrosion resistance, while PINNs integrate physical corrosion laws to enhance prediction accuracy [23][24]. These models are particularly effective for modeling multipoint corrosion or stress corrosion cracking (SCC) in pipeline steels [24].

These methods leverage large datasets, including experimental corrosion rates, field performance records, and simulated environmental conditions, to train models that predict material behavior with high precision [21][25].

2.3.2. Data Driven Prediction of Corrosion Resistance

Data driven approaches use extensive datasets to predict corrosion resistance, reducing reliance on time consuming lab tests. Alloy composition (elements like chromium, molybdenum, and nickel enhances resistance to pitting and SCC in stainless and duplex steels), environmental factors (variables such as CO₂/H₂S partial pressures, temperature, pH, and chloride concentration, critical for onshore pipeline corrosion), microstructural properties (grain size, phase distribution, and inclusions, which influence localized corrosion like pitting or MIC), mechanical properties (yield strength and toughness, which affect susceptibility to SCC under pipeline stresses). AI models process these inputs to predict outcomes like corrosion rates (mm/year), pitting depth, or time to failure under specific conditions. For instance, ANNs have been used to model the corrosion rate of carbon steel pipelines in CO₂ rich environments, achieving prediction errors below 5% compared to experimental data [17][19]. Similarly, SVMs and random forests can classify materials as resistant or susceptible to H₂S induced cracking, guiding the selection of duplex steels for sour service [21][22]. These predictions enable proactive material selection, optimizing for both performance and cost in onshore pipelines [23].

2.3.3. Case Studies

Several case studies highlight the efficacy of AI in material selection and corrosion prediction. High entropy alloys [25] applied deep learning to predict corrosion resistant compositions in high entropy alloys, identifying novel alloys with superior resistance to pitting and SCC. This approach is adaptable to pipeline steels, where AI can screen compositions for optimal chromium or molybdenum content. Corrosion rate forecasting in pipelines [17] developed an ANN model to predict internal corrosion rates in oil pipelines, using inputs like fluid composition and flow rate. The model achieved high accuracy in forecasting CO₂ corrosion, aiding the selection of carbon vs. stainless steels. Inhibitor efficiency modeling [19] used ANNs to predict the effect of corrosion inhibitors on CO₂ corrosion in carbon steels, demonstrating how AI can optimize material performance by factoring in chemical treatments. Stress corrosion cracking prediction [26] employed ANNs to predict time to failure for 304 stainless steels under SCC conditions, providing insights for selecting materials resistant to tensile stresses in pipelines. Microbiologically influenced corrosion (MIC) [27] applied AI to model MIC in offshore systems, adaptable to onshore pipelines, to predict degradation rates and select resistant materials like duplex steels. Vision based corrosion detection [28] used CNNs with micro aerial vehicles to detect corrosion on pipeline surfaces, supporting material selection by identifying early degradation patterns. Environmental assisted cracking, studies like [29] used ANNs and risk-based AI models to predict corrosion in equatorial or H₂S/CO₂ environments, informing material choices for harsh onshore conditions.

Implementing AI for material selection requires robust datasets, often sourced from experimental corrosion tests, field data, or simulations. Challenges include data quality (e.g., incomplete or noisy datasets) and computational complexity, particularly for DL models like CNNs or PINNs [24][30]. Feature selection is critical, as irrelevant inputs (e.g., non-corrosive environmental factors) can reduce model accuracy. Techniques like principal component analysis (PCA) or feature importance ranking in random forests help prioritize key parameters like alloy composition or chloride exposure [22]. Additionally, hybrid approaches combining AI with physical corrosion models (e.g., NACE standards) enhance reliability by grounding predictions in established science. AI assisted material selection leverages ML and DL to revolutionize the identification of corrosion resistant steels for onshore pipelines. By combining predictive power with case study validation, these methods enable engineers to select materials that balance performance, cost, and durability, addressing the unique corrosion challenges of CO₂, H₂S, MIC, pitting, and SCC.

Developing an AI-assisted framework for selecting corrosion resistant steels for onshore oil and gas pipelines involves a systematic approach to integrate material science, environmental data, and advanced computational techniques. The framework aims to optimize material selection by predicting performance under corrosive conditions, reducing costs, and enhancing pipeline longevity. This process encompasses defining comprehensive input parameters, training and validating AI models with robust datasets, and designing decision support systems to provide actionable recommendations for engineers. The framework leverages machine learning to bridge data gaps, improve prediction accuracy, and streamline decision making in complex, corrosive environments typical of onshore pipelines. The framework relies on a multidimensional set of input parameters to capture the factors influencing corrosion resistance and material suitability.

Alloy Composition: Key elements such as chromium, molybdenum, nickel, and carbon content significantly affect corrosion resistance. For example, higher chromium content enhances pitting resistance in stainless and duplex steels, quantified by the Pitting Resistance Equivalent Number [21][25].

Microstructure: Grain size, phase distribution (e.g., austenite, ferrite in duplex steels), and precipitate formation influence mechanical strength and susceptibility to stress corrosion cracking (SCC) or pitting [22]. Microstructural data can be derived from metallographic analysis or simulations.

Mechanical Properties: Yield strength, tensile strength, and hardness are critical for ensuring pipelines withstand operational stresses while resisting corrosion induced degradation [24]. For instance, duplex steels offer high yield strength for high pressure pipelines.

Corrosion Properties: Metrics like corrosion rate (mm/year), PREN, and susceptibility to specific mechanisms (e.g., CO₂ corrosion, H₂S induced cracking) are evaluated under simulated or field conditions [19]. These properties are often tested per standards like NACE TM0177 for sour service.

Service Environment: Environmental factors include pH, temperature, partial pressures of CO₂ and H₂S, chloride concentration, soil corrosivity, and flow conditions (e.g., velocity, turbulence). These parameters dictate corrosion severity and material performance in onshore settings [23][31]. For example, high CO₂ partial pressures exacerbate sweet corrosion, necessitating corrosion resistant alloys (CRAs).

These inputs are compiled from experimental data, field measurements, and industry databases, ensuring the framework accounts for real world variability in pipeline operations [30].

AI Model Training and Validation

AI model development for selecting corrosion resistant steels in onshore oil and gas pipelines involves selecting appropriate algorithms and training them on high quality datasets to predict material performance under corrosive conditions. Common machine learning approaches include artificial neural networks (ANNs), which excel at modeling complex, non-linear relationships between alloy composition and corrosion rates, and support vector machines (SVMs), which classify materials as suitable or unsuitable based on environmental thresholds [22]. Decision trees and random forests offer interpretable predictions, making them accessible for engineers less familiar with AI [27]. Physics informed neural networks (PINNs) enhance accuracy by incorporating physical laws, such as electrochemical kinetics and mass transport equations, reducing reliance on large datasets, a critical advantage for niche applications like onshore pipelines where data may be scarce [24]. For example, PINNs can model CO₂ corrosion rates by integrating Faraday’s laws with environmental data. Training datasets are sourced from laboratory experiments (e.g., autoclave tests for H₂S corrosion), field corrosion monitoring (e.g., weight loss coupons, electrochemical probes), and computational simulations (e.g., finite element modeling of stress corrosion cracking) [26], supplemented by public databases like NACE Corrosion Data or proprietary industry datasets. Model validation employs techniques like k-fold cross validation to ensure generalizability across diverse pipeline conditions, such as testing a model trained on carbon steel corrosion in CO₂ environments across varying temperatures and pH levels to confirm robustness [32]. Performance is quantified using metrics like mean squared error for regression tasks or F1 score for classification, with hyperparameter tuning (e.g., adjusting learning rates or ANN hidden layers) optimizing outcomes [33]. Data preprocessing, including handling missing values, normalizing inputs like PREN or corrosion rates, and addressing outliers, is essential for reliable predictions, while techniques like synthetic data generation or transfer learning mitigate data scarcity for rare corrosion scenarios, such as microbiologically influenced corrosion in specific soil types [34]. Figure 3 shows AI-assisted framework for selecting corrosion resistant for an onshore oil and gas pipelines.

Figure 3. AI-assisted framework for selecting corrosion resistant for an onshore oil and gas pipelines.

The integration of artificial intelligence (AI) into material selection for corrosion resistant steels in onshore oil and gas pipelines offers transformative advantages over traditional methods, but it also introduces unique challenges. This compares the strengths and weaknesses of AI driven approaches, such as machine learning (ML) and deep learning (DL), against conventional methods like performance testing and empirical modeling, while addressing methodological hurdles including data scarcity, generalizability, and interpretability.

2.5.1. Strengths of AI Methods

AI techniques, such as artificial neural networks (ANNs), support vector machines (SVMs), and ensemble learning, excel in processing complex, high dimensional datasets to predict corrosion behavior and material performance with high accuracy [35]. Unlike traditional methods, which rely on time consuming laboratory tests, AI models can rapidly analyze vast datasets, identifying patterns in alloy composition, environmental conditions, and corrosion rates that may elude empirical models [36]. For instance, ANNs have been used to predict CO₂ corrosion rates in carbon steels with errors below 5%, significantly faster than electrochemical testing [37]. AI’s scalability allows it to handle diverse pipeline conditions, from sour (H₂S rich) to sweet (CO₂ rich) environments, enabling proactive material selection for specific onshore settings [38]. Additionally, AI frameworks can integrate multi-objective optimization, balancing cost, durability, and corrosion resistance, as demonstrated in decision support systems for duplex steel selection [39]. These systems reduce human bias and streamline decision making compared to subjective cost benefit analyses.

2.5.2. Weaknesses of AI Methods

Despite their advantages, AI methods depend heavily on high quality, comprehensive datasets, which are often scarce for niche applications like onshore pipelines exposed to microbiologically influenced corrosion (MIC) or stress corrosion cracking (SCC) [40]. Unlike traditional methods, which are grounded in well-established physical principles (e.g., Faraday’s laws for corrosion kinetics), AI models, particularly deep learning, can act as “black boxes”, offering limited insight into the physical mechanisms driving predictions. This lack of interpretability can undermine trust among engineers accustomed to transparent empirical models like those in NACE MR0175 standards [41]. Furthermore, AI models trained on specific datasets (e.g., laboratory-based corrosion tests) may struggle to generalize to real world pipeline conditions, such as varying soil chemistries or fluctuating temperatures.

2.5.3. Strengths of Conventional Methods

Traditional approaches, such as immersion testing, electrochemical analysis, and empirical models, provide reliable, physics-based insights into corrosion behavior. For example, standards like NACE MR0175 offer clear guidelines for selecting materials in H₂S environments, ensuring compliance with industry regulations. These methods are well validated, with decades of field data supporting their use, such as corrosion rate predictions for carbon steels in CO₂ rich pipelines [13]. Their transparency allows engineers to trace the rationale behind material choices, making them suitable for critical applications where safety is paramount.

2.5.4. Weaknesses of Conventional Methods

Conventional methods are often labor intensive and time consuming, requiring extensive testing to evaluate materials under simulated pipeline conditions. For instance, pitting corrosion tests for stainless steels can take weeks to yield results [42]. Empirical models, while robust, rely on historical data that may not account for novel alloys or extreme environments, leading to conservative designs or overlooked risks. Subjectivity in interpreting test results or weighing cost benefit tradeoffs can also introduce inconsistencies, particularly when selecting between carbon and duplex steels for aggressive onshore settings [43].

2.5.5. Methodological Challenges for AI

Several methodological challenges impede the widespread adoption of AI in material selection for corrosion resistant steels in onshore oil and gas pipelines. Data scarcity poses a significant hurdle, as comprehensive datasets linking alloy composition, microstructure, and corrosion performance in specific environments, such as high chloride soils, are often limited. For instance, microbiologically influenced corrosion (MIC) data for onshore pipelines is frequently incomplete due to the complex interplay of microbial interactions, leading to overfitting in AI models and reduced reliability [44]. Generalizability is another critical issue, as AI models trained on controlled laboratory conditions may struggle to adapt to diverse field conditions, such as varying pH levels or H₂S partial pressures encountered in onshore pipelines [38]. While transfer learning and domain adaptation techniques are being explored to enhance generalizability, these approaches demand substantial computational resources, complicating their practical implementation [45]. Additionally, the interpretability of black box models, such as deep neural networks, remains a challenge, as these models obscure the causal relationships between inputs like chromium content and outputs like corrosion rate, making it difficult to justify material selections to stakeholders. Hybrid approaches, such as physics-informed neural networks, offer a promising solution by integrating physical laws to improve interpretability, as demonstrated in models predicting stress corrosion cracking (SCC) in stainless steels [46].

AI’s role in predictive corrosion modeling is rapidly expanding, driven by advancements in machine learning (ML) and deep learning (DL) techniques such as artificial neural networks (ANNs), support vector machines (SVMs), and ensemble methods [35][37]. These models excel at analyzing complex datasets, predicting corrosion rates, and identifying optimal material compositions for challenging environments like high chloride soils or H₂S-rich pipelines [25]. For example, ANNs have achieved prediction errors below 5% for CO₂ corrosion in carbon steels, far surpassing the speed of traditional electrochemical testing [37]. The emergence of hybrid AI-physics models, such as physics informed neural networks, further enhances predictive accuracy by incorporating corrosion kinetics, as demonstrated in studies modeling stress corrosion cracking (SCC) in stainless steels [46][47]. Additionally, AI-driven decision support systems are streamlining material selection by optimizing tradeoffs between cost, durability, and corrosion resistance, reducing reliance on subjective cost benefit analyses [17]. This trend reflects a broader shift toward data-driven materials science, with AI enabling proactive rather than reactive corrosion management strategies.

Despite these advancements, significant knowledge gaps hinder AI’s full adoption in pipeline material selection. A primary challenge is data scarcity, as comprehensive datasets linking alloy composition, microstructure, and corrosion performance in specific onshore environments (an example is high chloride soils or microbiologically influenced corrosion [MIC] settings) are limited. For instance, MIC data is often incomplete due to the complexity of microbial interactions, leading to overfitting and reduced model reliability [48]. Another gap is the lack of field validation for AI models, which are frequently trained on controlled laboratory data that may not capture the variability of real-world pipeline conditions, such as fluctuating pH or H₂S partial pressures [38]. While transfer learning and domain adaptation are being explored to improve generalizability, these techniques require significant computational resources and are not yet widely validated in pipeline applications [49]. Furthermore, the interpretability of black box models like deep neural networks remains a barrier, as their opaque decision-making processes can undermine stakeholder trust compared to transparent empirical models like those in NACE MR0175 [41][50]. These gaps highlight the need for robust datasets and field-tested AI frameworks to ensure reliability in practical applications [51].

The data generation in this strategy consists of two parts [52] the generation of input variables (environmental factors) and output variables (corrosion depth) for corrosion growth prediction. For the outliers in the corrosion dataset, multiple detection methods were used for data cleaning. For tabular data and oversampling, CTGAN, which introduces mode-specific normalization and training by sampling, is used to learn real data to generate new environmental factors. For the regression challenge, a corrosion growth model based on machine learning algorithms is applied to generate corrosion depths corresponding to the new environmental factors. The fake data are merged with the original data to form the synthetic dataset. Finally, the effectiveness of the proposed strategy is verified by analyzing the distribution and credibility of the synthetic dataset using various visualization methods and evaluation metrics. This technique can provide data support for future corrosion growth modeling, and this method can also be applied to data generation for regression problems in other fields.

The adoption of AI-assisted material selection frameworks promises more reliable and cost-effective solutions for onshore pipeline corrosion management. By rapidly analyzing vast datasets, AI models can identify corrosion resistant steels (e.g., duplex or stainless steels) tailored to specific environmental conditions, reducing the risk of pipeline failures due to CO₂, H₂S, or MIC [27][37]. AI-driven decision-support systems can rank materials based on multi-objective optimization, balancing corrosion resistance with lifecycle costs, which is particularly valuable for resource-constrained operators [21]. These systems also minimize human bias inherent in traditional methods, such as subjective interpretations of test results. Moreover, hybrid AI-physics models improve prediction reliability by integrating physical principles, as seen in CO₂ corrosion models achieving 10% higher accuracy than purely data-driven approaches [24]. In practice, these advancements could reduce maintenance costs and downtime, enhancing pipeline safety and longevity. However, practitioners must address data scarcity and model validation to ensure AI recommendations align with field performance and regulatory standards.

Materials science can also boost machine learning model explainability and performance in materials problems, as we have seen in the [53]. Physics informed machine learning is a blooming research field [54], instance according [55] topics that are related are general informed machine learning and theory guided data science Uncertainty quantification is another interesting field that can help ensure trust in ML model.