Development of Adaptive Neuro-Fuzzy Inference System for Assessing Industry Leadership in Accident Situations

Petroleum activity is characterized as a high-risk activity due to the probability of accidents with material and human losses. The leaders of this segment assume, besides the complex routine tasks, the challenge of making assertive decisions during an accident. This study aims to present an evaluation model of the Industry Leadership Index for Emergencies Situations (ILIE), using the Adaptive Neuro-Fuzzy System (ANFIS). The model was composed of 4 input variables, namely: knowledge, behavior, skill, and attitude; and one output variable, Industry Leadership. The data collection took place in petroleum production units in Brazil, with a sample of 151 respondents through the application of a survey. The observed data were treated in an Excel tabulator and used in the development of the ANFIS model. From this model, it was possible to carry out simulations to predict the impact, which the increase or decrease in the value of each input variable can influence the leader’s profile. The model performed satisfactorily in the Root of the Mean Square Error (RMSE) analysis, being 0.199 in data training and 1.217 in data verification. The results suggest that the ANFIS method can be successfully applied to establish a model to analyze industry leaders prepared for assertive responses in crisis scenarios.

The associate editor coordinating the review of this manuscript and approving it for publication was Yongquan Sun . examples of procedures with non-assertive answers, as well 66 as content to explore manners that should not be used and 67 those which could be used in cases of serious accidents, such 68 as the two recently cited [10], [15]. 69 This research presents a model for evaluating the Indus-70 try Leadership Index for Emergencies using the Adaptive 71 Neuro-Fuzzy System (ANFIS) [1]. The model presents four 72 input variables: Knowledge (K ), Skill (S), Behavior (B), 73 and Attitude (A), as well as an output variable, the Industry 74 Leadership Index for Emergencies (ILIE). The data were 75 collected using a Likert scale survey applied to managers, 76 supervisors, and engineers who work in the operational area. 77 The paper is organized into six sections. Section 1 presents 78 the research, the context, and the objectives. Section 2 79 discusses the literature review. Part 3 highlights the model 80 development process, the variables evaluated, and the mea-81 surement tools. Section 4 describes the materials and meth-82 ods. Section 5 presents results and discussions. And section 6 83 describes the conclusions, research contributions, and future 84 work. 86

87
There is a specific leadership for each organization or sit-88 uation, whether formal or informal, based on factors such 89 as knowledge, techniques, and behaviors. These charac-90 teristics help in the engagement, motivation, and coop-91 eration of his/her followers, besides the performances 92 according to individual and group skills, and experiences [8], 93 [14], [16]. 94 The accidents at the Three Mile Island nuclear power plant 95 and the Piper Alpha Oil Platform in 1979 and 1988, respec-96 tively, led the scientific community to discuss new concepts 97 about process safety in the industry, from procedures arising 98 from human factors, which interfere in the responses of lead-99 ers in a crisis. These events caused fatalities and economic 100 losses, and the lessons learned were not enough to prevent 101 accidents from recurring. It can be observed that failures 102 persist, and in this context, leadership is an approach little 103 discussed when it comes to industry leaders. Leading teams 104 in highly complex spaces, such as the petroleum industry, 105 requires deepening attributes and indicators that can assess 106 industry leadership for emergencies [17]. 107 In an industrial environment, personal and group charac-108 teristics impact decision-making in serious situations, among 109 these characteristics there are commitment, cooperation, 110 competence, and communication. Regarding leadership style, 111 other characteristics are highlighted, and they are: prac-112 tical knowledge, control, interpersonal relationship, inclu-113 sion, humbleness, determination, experience, technique, safe 114 behavior, and emotional intelligence [18], [19], [20].

115
There are principles to be met to attribute to the leader, 116 the effectiveness in the activities developed in the process 117 industry to maintain a safe environment. Communication, the 118 skill to make decisions, responsibility and self-recognition, 119 simplicity, and humbleness are indicators of attributes of 120 leadership prepared to face an unexpected event that can lead 121 to victims and property damage. These are also important 122 attributes for the expansion of the safety culture [21], [22] important to discuss, in addition to the characteristics, the cul-178 ture, the relationships, and both the leaders' behaviors and the led's. In GLOBE, the evaluation of the degree of confidence, 180 decision-making power, and innovation are elements inserted 181 in the case study of the consultation with experts and leaders 182 in this paper [38], [39], [40] 183 The methods for leadership evaluation have a self-184 assessment format with the application of surveys. These dis-185 cussions restrict the activities in an environment of sectorial 186 accidents, as well as behaviors and procedures that ensure the 187 leader's competence to act in these circumstances. The liter-188 ature consulted contributed to the identification of the vari-189 ables and indicators used for the development of the model 190 specifying the leader's activity area and how to respond to 191 emergencies.  Theoretical knowledge is part of learning concepts, princi-205 ples, and fundamentals, to identify characteristics and rules 206 according to what they represent for learning, with spe-207 cialized information and content. Procedural knowledge is 208 defined as that which corresponds to the exercise of the 209 activity and its specificity, a set of techniques, strategies, and 210 methods of performing the task [42], [43].

211
When the impact of a threat occurs during routine activ-212 ities, knowledge is not always isolated, leading to the best 213 decision and response to the event. There is a high probability 214 of initiating high-impact, low-visibility failures during the 215 progressive stress stage in an accident occurrence. Cognitive 216 functions are altered: memory, attention, logical reasoning, 217 perception, and decision, that is why they react with the inclu-218 sion of auxiliary memory through notes, documents, mind 219 maps, and safety systems, to achieve successful decisions 220 [13], [44].

221
The indicators were identified and subjected to analysis as 222 to the degree of importance for an industry sector leader in an 223 emergency, according to the presentation in Table 1.

225
It is possible to demonstrate and act with determination 226 and emotional control to achieve safe behavior. One of the 227 pillars of leadership is confidence, the values, and beliefs 228 that inspire conscious actions in environments vulnerable to 229 accidents [48]. 230 Inappropriate behaviors can be mapped by the leader's 231 psychological factors and the group's. It is observed that the 232 unconscious arises without warning, and preventive measures 233 VOLUME 10, 2022 cannot be triggered. When leadership has unique hazard con-   set. In an uncertain event, risk perception, creativity, and 269 resilience are part of this construct.

270
Enabled people keep track of their activities step by step, 271 with steps memorized and repeated to accomplish the tasks. 272 An individual may be able to manage a crisis, with the 273 development of a mind map. It is important not to rely on 274 automation, because of the unconscious forgetfulness that can 275 be part of a high-stress environment 276 Skill is not a substitute for the rule, they are complementary 277 attributes that allow the action to be performed, with the abil-278 ity to formulate the rules that are validated with the consensus 279 of pairs. Those who routinely experience the same practical 280 situations and adopt automatic behaviors, even repeating the 281 work rituals, cannot always be considered safe. Knowledge, 282 skill, and intuitive learning for high-stress situations must 283 come together to structure leadership suitable for quick and 284 assertive decisions in industry emergencies [23].

285
The leader mobilizes, aggregates, and encourages. If these 286 skill indicators, as well as creativity and organization, democ-287 ratize learning for impactful execution, in managing an emer-288 gency in the industry. Assertive behaviors and responses 289 influence driving stress control and cognitive memory 290 activation [42], [60]. Table 3 presents the variety of indicators that are associ-292 ated with the variable Skill. The fields of education, health, 293 politics, business, psychology, and administration discuss the 294 impact of this attribute, and correlations with other variables 295 are discussed in this paper. 296

297
The mental disposition related to the specific situation, and 298 the subject, comparing scenarios with customized reactions, 299 can be one of the definitions of attitude. According to 300 researchers, when faced with a situation that requires a deci-301 sion and no contest, attitude is a choice. This requires a sense 302 of balance so that the choice leads to the desired response.

303
For this, it is necessary to be a participative agent [9].

304
The outcome is a consequence of this choice, and the 305 tendency is the presence of an action conditioned to cog-306 nitive processing. Tolerance and emotion go hand in hand 307 with aspects related to command, courage, and positioning 308 in the face of adversity that involves risks and contingency in 309 emergencies [67]. [68], [69], [70]. The ANFIS is an intelligent system with the learning 328 capability of neural network architecture. Thus, the ANFIS 329 simultaneously processes linguistic variables and learns from 330 the environment in which it is inserted.

331
It is an architecture that adjusts the parameters of a fuzzy 332 set of inputs and outputs, and the observed data is used in 333 the process of developing the System and brings together two 334 types of modeling of fuzzy sets, based on the Takagi-Sugeno 335 fuzzy inference system. This model was chosen because it 336 interprets the system simpler [1], [77].

337
The Takagi-Sugeno fuzzy inference system uses a mapping 338 system for each fuzzy IF-THEN rule output. This function 339 maps the input and output of the rule from a combination of 340 the inputs. The fuzzy rules from the SUGENO model do not 341 use the fuzzy set, but rather a mathematical function from the 342 inputs. The most common rules format is: The input of this model (Fig. 1) is composed of two vari-355 ables (x and y) and an output variable (F). The first layer (1) 356 represents the fuzzification stage, and each input node (Ï) 357 is 'an adaptive node, i.e., the degree of adherence to the 358 linguistic term, calculated based on the premise of each rule, 359 represented in A 1 , A 2 , B 1 , and B 2 . In the second layer (2), each 360 node ( ) represents a rule. At this step, the result is calculated 361 which will determine the consequent degree of the rule that 362 will be achieved. The impact of each rule is represented by 363 W 1 and W 2 . In the third layer (3), the normalized value of the 364 activation degree of each rule is calculated. Each node in this 365 layer is identified with the letter N, and the normalized output 366       with 145 (96%) men, and 6 (4%) women. Among the respon-433 dents, 89% had high school, and 11% had postgraduation.

434
Twenty-three forms were applied on-site at one of the 435 petroleum production units in the Northeast of Brazil.

436
After this test, 177 forms were sent by e-mail, distributed 437 among 5 petroleum production units in Brazil, 160 of which 438 were returned, and 9 blank surveys were excluded.

439
To verify the consistency of the survey as to its reliability, 440 Cronbach's alpha coefficient was used, with a criterion of 441 0.7 to 0.9 for being considered adequate. This is one of 442 the statistical procedures for measuring internal consistency, 443 which refers to the degree to which the survey items are 444 correlated with each other and with the research result.  Considering that, in the literature, no studies were found 451 for the evaluation of industry leadership for emergencies, the 452 evaluation criteria were based on a scale from 0 to 100% in the 453 questionnaire applied in the case study, from the discussions 454 with specialists. The scale defined by the specialists considered unsatisfactory values below 59%, regular from 60% to 456 69%, and satisfactory above 69%.

507
The defined stopping training criteria were the error toler-508 ance equal to 10 − 7 and the number of epochs equal to 150. 509 After 6 tests, the number of epochs with the lowest RMSE 510 was identified, according to the type of function of Gaussian 511 relevance. Table 6 shows information from ANFIS after train-512 ing the input data in the Fuzzy Inference System.

514
The Fuzzy Inference System tests were performed with the 515 observed data separated three times: training, test, and vali-516 dation, according to the linguistic variables, to measure per-517 formance values: unsatisfactory, regular, and satisfactory. The 518 Gaussian inference with Activation Function represented by 519 Equation 3 was used in the system [83]. Knowledge.

546
Having made the definitions presented in this section, the 547 model results were analyzed and discussed.

549
The study was divided into two stages, the first being the data  System, to observe the impact of each variable on the industry 553 leader's performance for an emergency.

554
The Cronbach's Alpha value was 0.7, which represents the 555 result reliability, considered adequate for studies presenting 556 subjective data and a questionnaire based on the Likert scale 557 used in this study [84].

558
In the Excel Tabulator, on a scale from 0 to 100, the index 559 for the variable Knowledge was 80%, Behavior 73%, Skill 560 66%, and Attitude 61%. Given the criteria adopted by the 561 experts, the variables Knowledge and Behavior were consid-562 ered of satisfactory values and the variables Skill and Attitude 563 obtained regular ones, while the leadership index of 70% 564 was considered adequate or satisfactory as a parameter for 565 evaluating an industry leader in emergencies.

566
The model showed the lowest RMSE at the 150th epoch, 567 outperforming the other models in the tests performed. The 568 results obtained by the ANFIS model showed that, when 569 reaching the defined stopping criteria of error tolerance and 570 number of epochs, the RMSE values were 0.199 in the data 571 training, and 1.217 in the data validation. The model per-572 formance was satisfactory considering that the RMSE value, 573 the smaller the difference between the estimated and the real 574 values [85], [86].

575
According to the Rule Viewer results (Figure 7), the influ-576 ence of input variables in determining the output index can 577 be measured. By moving the red line in the center of each 578 input variable to its maximum value, it is possible to verify 579 the positive impact of this variable on the leadership profile. 580 The Rule Viewer results were compared to the calculations 581 performed in the Excel Tabulator to present the percentage 582 errors between real data and the simulation through ANFIS. 583 In the Neuro-fuzzy Designer application, the model results 584 can be seen in the Rule Viewer and the Surface Viewer. Based 585 on the Rule Viewer results, the percentage errors considering 586 the real values and the simulated values were calculated. The 587 percentage errors were calculated using the model input data 588 (real or observed values), and the results of each variable in 589 the Rule Viewer (simulated values). The percentage errors 590    In this sense, the paper contributes with new content, that 672 specifically addresses the evaluation of the industry leader.

673
In addition to emerging further discussion and theories on the 674 topic of this paper. Assertive responses in a crisis scenario 675 also depend on the execution of procedures by the leader and 676 teams that can contingency an accident.

677
For future work, this research should be applied to other 678 organizations with similar characteristics to the oil and gas 679 industry for validation of the tool. Reliability Analysis to Prevent Losses in Industrial Processes. 955 P. P. S. CARVALHO was born in Bahia. She 956 received the degree in production engineering 957 from Santa Cruz State University, Bahia,Brazil,958 in 2009, and the master's degree in industrial engi-959 neering from the Polytechnic School, Federal Uni-960 versity of Bahia, in 2013, where she is currently 961 pursuing the Ph.D. degree. She has experience 962 in production engineering, focusing on logistics, 963 supply chain management, production planning 964 and control, ergonomics, hygiene, and work safety. 965 She is currently an Assistant Professor with the Production Engineering 966 Course, State University of Santa Cruz. She has served as a member for the 967 Evaluation Committee of the V and VI Brazilian Congress of Production 968 Engineering. She has also served as a Reviewer for IEEE ACCESS journal. issues, such as human aspects of process control, 990 operation production systems, operational reliabil-991 ity, worker profiles in character and personality to human errors, skills to 992 reduce process losses, management and techniques for risk control, human 993 and organizational factors for accident avoidance, and diagnosis of human 994 factors to increase reliability.