breakpilot-compliance/ai-compliance-sdk/internal/iace/narrative_parser.go

package iace

import (
	"regexp"
	"strconv"
	"strings"
)

// ComponentMatch represents a component detected from narrative text.
type ComponentMatch struct {
	LibraryID  string   `json:"library_id"`
	NameDE     string   `json:"name_de"`
	MatchedOn  string   `json:"matched_on"`  // The keyword that triggered the match
	Tags       []string `json:"tags"`
	Confidence float64  `json:"confidence"`
}

// EnergyMatch represents an energy source detected from narrative text.
type EnergyMatch struct {
	SourceID  string  `json:"source_id"`
	NameDE    string  `json:"name_de"`
	MatchedOn string  `json:"matched_on"`
	Value     string  `json:"value,omitempty"` // e.g., "20000 kN", "400 V"
	Severity  int     `json:"severity"`        // Derived severity 1-5
}

// TechSpec represents an extracted technical specification.
type TechSpec struct {
	Value float64 `json:"value"`
	Unit  string  `json:"unit"`
	Raw   string  `json:"raw"`
}

// ParseResult contains all entities extracted from a machine narrative.
type ParseResult struct {
	Components        []ComponentMatch `json:"components"`
	EnergySources     []EnergyMatch    `json:"energy_sources"`
	LifecyclePhases   []string         `json:"lifecycle_phases"`
	Roles             []string         `json:"roles"`
	CustomTags        []string         `json:"custom_tags"`
	TechSpecs         []TechSpec       `json:"tech_specs"`
	Confidence        float64          `json:"confidence"`
	OperationalStates []string         `json:"operational_states,omitempty"`
	StateTransitions  []string         `json:"state_transitions,omitempty"`
}

// techSpecPattern matches numeric values with engineering units.
var techSpecPattern = regexp.MustCompile(`(\d[\d.,]*)\s*(kN|Tonnen|tonnen|kJ|kW|MW|V|kV|Hz|°C|bar|mm|m³/h|/min|U/min|rpm|m/s)`)

// lifecycleKeywords maps German text patterns to lifecycle phase IDs.
var lifecycleKeywords = map[string]string{
	"betrieb":            "normal_operation",
	"normalbetrieb":      "normal_operation",
	"automatikbetrieb":   "auto_operation",
	"einricht":           "setup",
	"umruest":            "changeover",
	"werkzeugwechsel":    "changeover",
	"wartung":            "maintenance",
	"instandhalt":        "maintenance",
	"instandsetz":        "repair",
	"reinig":             "cleaning",
	"transport":          "transport",
	"montage":            "assembly",
	"installation":       "assembly",
	"inbetriebnahme":     "commissioning",
	"ausserbetriebnahme": "decommissioning",
	"demontage":          "disposal",
	"entsorgung":         "disposal",
	"reparatur":          "repair",
	"stoerungsbeseitig":  "fault_clearing",
	"stoerung":           "fault_clearing",
	"fehlersuche":        "fault_clearing",
	"klemm":              "fault_clearing",
	"blockier":           "fault_clearing",
	"stau":               "fault_clearing",
}

// roleKeywords maps German text patterns to role IDs.
var roleKeywords = map[string]string{
	"bedienpersonal":  "operator",
	"bediener":        "operator",
	"werker":          "operator",
	"einrichter":      "setup_personnel",
	"instandhalt":     "maintenance_tech",
	"wartungspersonal": "maintenance_tech",
	"elektrofachkraft":"electrical_tech",
	"besucher":        "visitor",
	"fremdfirma":      "contractor",
	"reinigungspersonal": "cleaning_staff",
	"aufsichtsperson": "supervisor",
	"programmierer":   "programmer",
	"auszubildend":    "trainee",
	"leiharbeiter":    "temp_worker",
}

// operationalStateKeywords maps German text patterns to operational state IDs.
var operationalStateKeywords = map[string]string{
	"hochfahren":          "startup",
	"anlauf":              "startup",
	"anfahren":            "startup",
	"referenzfahrt":       "homing",
	"referenzpunkt":       "homing",
	"automatikbetrieb":    "automatic_operation",
	"automatisch":         "automatic_operation",
	"handbetrieb":         "manual_operation",
	"manuell":             "manual_operation",
	"tippbetrieb":         "manual_operation",
	"teach":               "teach_mode",
	"einrichtbetrieb":     "teach_mode",
	"programmier":         "teach_mode",
	"wartung":             "maintenance",
	"instandhaltung":      "maintenance",
	"reinigung":           "cleaning",
	"not-halt":            "emergency_stop",
	"nothalt":             "emergency_stop",
	"notabschaltung":      "emergency_stop",
	"wiederanlauf":        "recovery_mode",
	"wiederinbetriebnahme":"recovery_mode",
	"quittier":            "recovery_mode",
}

// stateTransitionKeywords maps keyword combinations to state transitions.
var stateTransitionKeywords = map[string]string{
	"unerwarteter wiederanlauf": "maintenance→automatic_operation",
	"wiederanlauf nach not":     "emergency_stop→recovery_mode",
	"automatischer anlauf":      "startup→automatic_operation",
	"betriebsartwechsel":        "manual_operation→automatic_operation",
}

// ParseNarrative extracts components, energy sources, lifecycle phases,
// roles, and tags from a machine description text. Fully deterministic,
// no LLM required.
// machineType is optional — if provided, keywords with MachineTypes
// restrictions are only matched when the machine type is in the list.
func ParseNarrative(text string, machineType ...string) ParseResult {
	result := ParseResult{}
	if text == "" {
		return result
	}

	// Normalize text
	lower := strings.ToLower(text)
	lower = strings.ReplaceAll(lower, "ä", "ae")
	lower = strings.ReplaceAll(lower, "ö", "oe")
	lower = strings.ReplaceAll(lower, "ü", "ue")
	lower = strings.ReplaceAll(lower, "ß", "ss")

	// 1. Extract technical specifications
	result.TechSpecs = extractTechSpecs(text)

	// 2. Match keywords → components + energy + tags
	dictionary := GetKeywordDictionary()
	compLib := GetComponentLibrary()
	compMap := make(map[string]ComponentLibraryEntry)
	for _, c := range compLib {
		compMap[c.ID] = c
	}

	seenComponents := make(map[string]bool)
	seenEnergy := make(map[string]bool)
	tagSet := make(map[string]bool)

	// Resolve machine type for filtering
	var mType string
	if len(machineType) > 0 {
		mType = machineType[0]
	}

	for _, entry := range dictionary {
		// Skip keywords restricted to other machine types
		if len(entry.MachineTypes) > 0 && mType != "" {
			matched := false
			for _, mt := range entry.MachineTypes {
				if mt == mType {
					matched = true
					break
				}
			}
			if !matched {
				continue // This keyword is for a different machine type
			}
		}

		for _, kw := range entry.Keywords {
			kwNorm := strings.ToLower(kw)
			kwNorm = strings.ReplaceAll(kwNorm, "ä", "ae")
			kwNorm = strings.ReplaceAll(kwNorm, "ö", "oe")
			kwNorm = strings.ReplaceAll(kwNorm, "ü", "ue")
			kwNorm = strings.ReplaceAll(kwNorm, "ß", "ss")

			if strings.Contains(lower, kwNorm) {
				// Add components
				for _, cid := range entry.ComponentIDs {
					if !seenComponents[cid] {
						seenComponents[cid] = true
						comp := compMap[cid]
						result.Components = append(result.Components, ComponentMatch{
							LibraryID:  cid,
							NameDE:     comp.NameDE,
							MatchedOn:  kw,
							Tags:       comp.Tags,
							Confidence: 0.8,
						})
						// Add component tags
						for _, t := range comp.Tags {
							tagSet[t] = true
						}
					}
				}
				// Add energy sources
				for _, eid := range entry.EnergyIDs {
					if !seenEnergy[eid] {
						seenEnergy[eid] = true
						result.EnergySources = append(result.EnergySources, EnergyMatch{
							SourceID:  eid,
							NameDE:    eid, // Will be enriched by caller
							MatchedOn: kw,
						})
					}
				}
				// Add extra tags
				for _, t := range entry.ExtraTags {
					tagSet[t] = true
				}
				break // First keyword match is enough per entry
			}
		}
	}

	// 3. Derive energy from tech specs
	for _, spec := range result.TechSpecs {
		deriveEnergyFromSpec(spec, &result, seenEnergy, tagSet)
	}

	// 4. Extract lifecycle phases
	phaseSet := make(map[string]bool)
	for kw, phase := range lifecycleKeywords {
		kwNorm := strings.ReplaceAll(kw, "ä", "ae")
		kwNorm = strings.ReplaceAll(kwNorm, "ö", "oe")
		kwNorm = strings.ReplaceAll(kwNorm, "ü", "ue")
		if strings.Contains(lower, kwNorm) {
			if !phaseSet[phase] {
				phaseSet[phase] = true
				result.LifecyclePhases = append(result.LifecyclePhases, phase)
			}
		}
	}

	// 5. Extract roles
	roleSet := make(map[string]bool)
	for kw, role := range roleKeywords {
		if strings.Contains(lower, kw) {
			if !roleSet[role] {
				roleSet[role] = true
				result.Roles = append(result.Roles, role)
			}
		}
	}

	// 6. Extract operational states
	stateSet := make(map[string]bool)
	for kw, state := range operationalStateKeywords {
		kwNorm := strings.ReplaceAll(kw, "ä", "ae")
		kwNorm = strings.ReplaceAll(kwNorm, "ö", "oe")
		kwNorm = strings.ReplaceAll(kwNorm, "ü", "ue")
		if strings.Contains(lower, kwNorm) {
			if !stateSet[state] {
				stateSet[state] = true
				result.OperationalStates = append(result.OperationalStates, state)
			}
		}
	}

	// 7. Extract state transitions
	transSet := make(map[string]bool)
	for kw, trans := range stateTransitionKeywords {
		if strings.Contains(lower, kw) {
			if !transSet[trans] {
				transSet[trans] = true
				result.StateTransitions = append(result.StateTransitions, trans)
			}
		}
	}

	// 8. Collect all tags
	for t := range tagSet {
		result.CustomTags = append(result.CustomTags, t)
	}

	// 9. Calculate overall confidence
	if len(result.Components) > 0 {
		result.Confidence = float64(len(result.Components)) / 15.0 // Normalize to ~1.0 for 15 components
		if result.Confidence > 1.0 {
			result.Confidence = 1.0
		}
	}

	return result
}

// extractTechSpecs finds numeric values with engineering units in the text.
func extractTechSpecs(text string) []TechSpec {
	matches := techSpecPattern.FindAllStringSubmatch(text, -1)
	var specs []TechSpec
	for _, m := range matches {
		valStr := strings.ReplaceAll(m[1], ".", "")
		valStr = strings.ReplaceAll(valStr, ",", ".")
		val, err := strconv.ParseFloat(valStr, 64)
		if err != nil {
			continue
		}
		specs = append(specs, TechSpec{
			Value: val,
			Unit:  m[2],
			Raw:   m[0],
		})
	}
	return specs
}

// deriveEnergyFromSpec maps technical values to energy sources and severity tags.
func deriveEnergyFromSpec(spec TechSpec, result *ParseResult, seen map[string]bool, tags map[string]bool) {
	switch {
	case (spec.Unit == "kN" || spec.Unit == "Tonnen" || spec.Unit == "tonnen") && spec.Value > 100:
		addEnergy(result, seen, "EN01", spec.Raw)
		tags["high_force"] = true
		if spec.Value > 1000 {
			tags["crush_point"] = true
		}
	case (spec.Unit == "V" || spec.Unit == "kV"):
		if spec.Value >= 400 || spec.Unit == "kV" {
			addEnergy(result, seen, "EN05", spec.Raw)
			tags["high_voltage"] = true
		} else if spec.Value >= 50 {
			addEnergy(result, seen, "EN05", spec.Raw)
			tags["electrical_part"] = true
		}
	case spec.Unit == "°C" && spec.Value > 60:
		addEnergy(result, seen, "EN06", spec.Raw)
		tags["high_temperature"] = true
		if spec.Value > 100 {
			tags["thermal_accumulation"] = true
		}
	case spec.Unit == "bar" && spec.Value > 10:
		addEnergy(result, seen, "EN07", spec.Raw)
		tags["high_pressure"] = true
	case (spec.Unit == "kW" || spec.Unit == "MW") && spec.Value > 1:
		addEnergy(result, seen, "EN02", spec.Raw)
		tags["rotating_part"] = true
	case (spec.Unit == "/min" || spec.Unit == "U/min" || spec.Unit == "rpm") && spec.Value > 100:
		addEnergy(result, seen, "EN02", spec.Raw)
		tags["rotating_part"] = true
		if spec.Value > 500 {
			tags["high_speed"] = true
		}
	case spec.Unit == "kJ" && spec.Value > 10:
		addEnergy(result, seen, "EN03", spec.Raw)
		tags["stored_energy"] = true
	}
}

func addEnergy(result *ParseResult, seen map[string]bool, id, matchedOn string) {
	if !seen[id] {
		seen[id] = true
		result.EnergySources = append(result.EnergySources, EnergyMatch{
			SourceID:  id,
			MatchedOn: matchedOn,
		})
	}
}